Skip to content

Silinx ISE 15.22.0

Choose a tag to compare

@github-actions github-actions released this 06 Nov 09:55
· 1356 commits to main since this release

What's new in 15.22.0

nextpnr is now Silinx's one place & route engine, design links open Silinx on a design for lecture slides, imported
projects are screened and trusted only by you, and every action of the interface is a tool of the MCP server.

In short:

  • One place & route engine, nextpnr: the Silinx toolchain places and routes with nextpnr on the Spartan-3E and the
    Spartan-6, in the browser too; Silinx's own placers and routers are gone. On the Spartan-3E it takes no more slices
    than they did.
  • Design links: Copy Link… opens Silinx on a design, a module, a test bench's simulation or an ASM chart's, state
    diagram's or Petri net's Simulate panel, with the pins and, if you choose, the current view (the open tabs, the split,
    the panels); a link's design can open beside your own project.
  • Imports screened: an imported project (Silinx or Xilinx ISE, a .zip, a link, the MCP server) and every file added
    to one are checked before Silinx reads them, and an imported project is untrusted until you trust it.
  • Simulate an ASM chart: a Simulate panel in the ASM chart editor, as the state diagram's.
  • Show … Beside: right-click a tab in Design or Simulate to show any related view beside it (a drawing's HDL, an HDL
    source's translation, the waveform's source).
  • Tabs in your order: drag a tab left or right to reorder it, as in a browser.
  • The editors, one layout: errors and warnings in the console (each check brings its tab to the front), a drawing's
    counts at the right of its toolbar, no hint or status lines, the read-only notice at the toolbar's right end; the ⋮
    menus of the simulator and the Board Emulator alike; Set as Top Module on an instantiated module shows it as the top.
  • I/O Pin Planning on the board: a Board view draws the board, its ports dragged onto the pins.
  • The ISE style interface has the Silinx one's themes, and Project ▸ Add to Project….
  • The MCP server does what the interface does: new sources and wizards, settings, renames, conversions, drawings,
    simulations of every model, forced signals, C simulation and co-simulation, imports and exports; the open project in
    the browser follows what it changes.
  • Silinx HLS: pointers into arrays, sizeof constants, named constants in pragmas and cases, RTL blackboxes on the
    boards.
  • Spartan-6: micro80 builds on the Nexys3 and the Atlys, and the examples build with ISE with no warnings.
  • The Board Emulator's VGA monitor no longer says "No VSYNC" on a slow computer while the first frames are still being
    simulated.

In detail:

  • Two UI tests that failed in the release's CI (#2906, #2907). The ISE style theme's test measures the element
    shown, as WebKit keeps the old style of one under display: none (the left pane's hidden pages); the handwritten
    expressions' test waits for the pad's first drawing before it reads its view.
    • Tests: test/ui/silinx.test.js, test/ui/ink.test.js.
  • No VSYNC only once 100 ms of the design ran with no frame (#2898). The monitor said its warnings once they had
    lasted a second of real time; on a slow machine the first two VSYNC pulses (about 33 ms of the design's time) may take
    longer, and it said "No VSYNC: the monitor finds no frames" for a design that has them. core/vga.js now waits for
    100 ms of the design's time, as for a VSYNC that stopped.
    • Tests: test/vga.test.js (no warning after one frame, the warning once 100 ms ran with no VSYNC).
  • The checks the release's CI found broken (#2897). The browser edition's flow asks the family's profile for the
    Spartan-6's nextpnr bundle (openFlow.bitDb), not the family's name; the tab order's Escape handler is onEscape, no
    longer taken for an HTML escape of its own; the ASM chart's simulated path's class is a listed safe shape. The tests
    follow a document tab's Show … Beside (the long press's menu, the interfaces' parity: the ISE style's View menu, the
    Silinx interface's tab menu), and on Windows compare cc's output with LF line ends and retry removing their scratch
    folder.
    • Tests: test/family-conditions.test.js, test/web-dom.test.js, test/web-markup-interpolations.test.js,
      test/ui/long-press.test.js, test/ui/interface-parity.test.js, test/c-sizeof-const.test.js, test/server-api.test.js.
  • A Silinx project imported in the browser keeps constraints/top.ucf (#2894): the browser's import of a Silinx
    export (a design's link, a .zip) left out a file at the path of an empty project's default constraints,
    constraints/top.ucf, the I/O Pin Planning's and most examples': the pins were lost and the file reported missing.
    Every file of a Silinx export is now kept (web/js/project-store.js projectFromZip). Tests: a regression test, a
    link's round trip of every example with constraints in the browser's store, the MCP's project_link / open_link
    (the project only, a module's link, the emulator's: the same .ucf, board, device and top, implemented with the
    same pinout and warnings) and the page's (the emulator's every port on its pin, Generate Programming File with no
    warnings).
  • The remove-module UI test follows #2872 (#2892): a drawing's badge shows the row's menu, the same as its module
    name's, with the drawing's commands in it; the test asserted the badge's old short menu.
  • The documents' tabs reordered by dragging (#2887): web/js/tab-order.js (in place of split-view.js's drag to
    the other side): a slot of the tab's width marks where it lands while the other tabs make room; Escape or a drop
    outside the rows cancels; a click still selects and the close button still closes; pointer events (a finger or a
    pen too). A focused tab moves with Alt+Shift+← / → (Ctrl+Shift+PageUp / PageDown where the browser leaves them),
    named in its tooltip and in Help ▸ Keyboard Shortcuts. The order is the documents' (each task's, Design's and
    Simulate's), kept with the open documents and restored in it, each side's in a split; Move to the Other Side and
    Show … Beside put a document at its new row's end. Portuguese; both interfaces; view-only, no MCP tool.
  • The view in a design's link (#2886): Copy Link…'s Include the current view (checked) adds ui= to the link: the
    open documents' reopen keys, the one in front, the split's right side, the left page, the Design view and the Silinx
    interface's task and arrangement (core/design-link.js encodeUi). Opening the link reopens them as a reload does
    (web/js/view-state.js applyViewState), then the link's own view; the layout is for that page only, over the
    viewer's kept one (web/js/tasks.js applyLayout), which a change of theirs keeps. Untrusted as the rest (#2829):
    only documents of the known kinds naming the project's files and modules, known pages, views and tasks, booleans,
    at most 40 documents and 4000 characters; what is left out is said in the console.
  • app.js's "where the user was" in a module of its own (#2886, #405): the open documents' state kept per project
    (#160, #163: saveViewState, restoreViewState, closeAllDocs and their reopening table) moved as it was to
    web/js/view-state.js, app.js re-exporting what moved; no change of behaviour. app.js from 998 lines to 939, room
    for a link to carry the view (#2886).
  • A link's board emulator emulates the top module (#2884): Copy Link… with Board Emulator let any module be
    chosen, and the link emulated that module, whose ports have none of the constraints' pins: the board showed nothing
    connected. Its Module list is now disabled on the top module, the link names no module, a link opens the emulator
    on the project's top, and a top= other than the top is refused with the reason in the console
    (core/design-link.js viewProblem, the MCP tools so too).
  • Copy Link to This Module… on a test bench (#2883): Copy Link… given a test bench opens on the simulation with
    that test bench chosen, and the link names it (view=simulate&top=<test bench>); the Silinx interface's test bench
    menu (Tasks ▸ Simulate's hierarchy and SSim's tree, web/js/tasks-sim-tree.js) has the item, as the ISE style's
    Simulation view has.
  • Copy Link… and a link's top= take a C module (#2882): a module written in C (Silinx HLS) was refused, "The
    design has no module": core/design-link.js read the design's modules from its HDL and its drawings only. Its C / C++
    sources' modules (their top function's, as the Design pane's) are now among them.
  • A design's link into an ASM chart's simulation (#2881): view=try is valid for a module drawn as an ASM chart
    (core/design-link.js TRY_DRAWING); the link opens the chart and its simulation panel through the ASM editor's
    showPanel('sim'), as the state diagram's and the Petri net's; until the editor has it, the chart opens and the
    console says its simulation is not in this version yet. Copy Link to This Module… chooses it; the MCP tools say so.
  • The ASM chart's Simulate panel (#2880): core/asm-sim.js simulates a chart cycle by cycle with its generated
    HDL's semantics (Verilog's widths, the every-cycle blocks before the state's assignments, the synchronised inputs,
    the synchronous or asynchronous reset), checked against the VHDL and the Verilog simulated, on the examples' charts
    and on edge cases (test/asm-sim.test.js); web/js/asm-sim.js is its panel (foldable, over the canvas on a narrow
    screen), sharing its inputs' controls and its trace with the state diagram's (web/js/sim-panel.js); a chart with
    errors is not simulated, its errors in the console; the editor's showPanel('sim') and panel open it and say
    whether it is open (a design link's view=try, #2881). The MCP server's simulate_drawing steps an ASM chart or a state diagram as their
    panels do. In both interfaces and in Portuguese (test/ui/asm-sim.test.js).
  • A link's simulation of a chosen test bench (#2878): Copy Link…'s Module list follows Open in: the simulation's
    lists the test benches (the modules of the files with the Simulation association, the project's first and
    chosen), the others the design's modules, no test bench. view=simulate&top=<test bench> simulates that test
    bench (with no top=, the project's); core/design-link.js viewProblem refuses a top= that is not a test bench
    with simulate, and a test bench with another view (benchNames); the MCP server's project_link and
    open_link so too.
  • The top always a root of the Implementation view (#2877): the hierarchy's roots are the top module, first, then
    the modules no module of the view instantiates (implRoots, core/elaborate.js, read by the Design pane's
    viewRoots); a top another module instantiates was an instance row only, with no top icon nor style. Its other
    appearances under other modules are plain instance rows; setting the top back on the parent restores the tree. The
    processes, Check Syntax, synthesis and the MCP server's set_top / project_info work on the same top.
  • A link's Try it found the drawing whatever its name's case (#2876): Copy Link… and view=try refused a module
    drawn as a schematic whose name differed in case from the module's (the HDL's VHDL names are in lower case, a
    drawing's as it was typed): "is not a schematic, a state diagram or a Petri net". A module and its drawing now match
    whatever their case, or through the drawing's generated HDL file; a link's top= is checked so too.
  • Open from a Link: Open as <name>_N (#2875): a project of the link's design's name that differs from it is
    replaced only on Replace (#2826); now the question's first answer, the primary, opens the link's design as a new
    project with the first free suffix (_2, _3, …; core/design-link.js freeName), the student's project kept,
    the link's view and module opened in the new one; Keep Mine and Replace as before. In Portuguese.
  • The Design pane's module menu the same everywhere (#2872): every badge of a module's row shows the row's menu
    (the module's, or a drawing's own for a drawing with no HDL file), no longer a short menu of the drawing's (#220).
    The drawing's commands joined the module's menu (Convert to State Diagram (FSM)… for an ASM chart, Convert to ASM
    chart for a state diagram, View State Machine (FSM) for a Petri net, Export as ISE Schematic (.sch)… for a
    schematic: web/js/drawings.js drawingCommands, in place of syncedViewMenu); its Open opens the drawing, else the
    HDL, as a double-click does.
  • A link's module selected (#2871): a link's top= selects its module's row in the Design pane (its parents
    expanded, scrolled to) and, with no view, opens its source as a double-click does (its drawing, else its HDL); with
    a view, the module is selected and the view opens as before. The Design pane's module rows carry their module's name
    (data-module). Copy Link to This Module… always puts its module in the link, the project's top too.
  • A design's link into a state diagram's or a Petri net's simulation (#2869): view=try is valid for a module
    drawn as a schematic, a state diagram or a Petri net (core/design-link.js tryDrawing): the link opens the drawing
    and its Simulate panel (the Petri net editor's showPanel, as the state diagram's); any other module is refused
    with the reason in the console. Copy Link to This Module… chooses it for those three; the dialog's option reads
    "Try it (a drawing's simulation)"; the MCP server's project_link and open_link describe it so.
  • A ⋮ menu's choices as menu items (#2846): toolbar-menu.js has checkable items for every viewer: menuChoice (a
    section title and its options, ✓ on the current one, an option that can't be chosen disabled with its reason as its
    tip) and menuCheck (a setting on or off), styled once with the other menus' items (.tb-menu-item), and ↑ / ↓ /
    Home / End move among a ⋮ menu's items. The Board Emulator's Engine, When hidden and Show (its modules) use them,
    its own menu styles (.emu-menu-row) gone; docs/UI-GUIDELINES.md: a ⋮ menu holds menu items, never form fields.
  • The local server watches the open project (#2843): server/project-watch.js looks at the folder of every
    project a page has open (its files, build/reports.json, build/silinx-status.json) and tells the page what
    another process changed (GET /api/projects/:p/events, server-sent events read through fetch, the token in a
    header); the server's own changes (a request of the page that may write, a run of its own) are not told back.
    The page (web/js/project-watch.js) reloads the project, an open document with no changes of its own (a deleted
    one closed), the processes' marks from a run made outside it, and the Design Summary; a document changed in the
    editor too asks "File Changed on Disk" (Keep the Editor's / Load from Disk). The browser edition has no events.
  • The waveform viewer's ⋮ menu (#2839): the simulator's toolbar ends with the documents' ⋮ menu
    (web/js/toolbar-menu.js), at the right of its first line, its place kept free whatever wraps (in the Tasks
    interface after the Model field); Save Wave Configuration… and Load Wave Configuration… are its entries, with their
    tooltips and Portuguese labels, and their two buttons leave the toolbar. The shared menu now works from the keyboard,
    in every document that has it: Enter or Space on ⋮ opens it on its first entry (↓ and ↑ too), ↑ / ↓ / Home / End move
    through its entries, Esc closes it back on ⋮; it is a menu for screen readers. A configuration's rows from and onto
    a design's signals are core/wcfg.js's (wcfgSignal, placeWcfg), shared by the wave window and the MCP.
    • MCP (#2674): save_wave_config writes a .wcfg of the simulation (the test bench's signals by default, or the
      rows given: signals with their radix, dividers, groups; the markers) next to its source, replacing a file only with
      overwrite; load_wave_config reads one onto the design, its signals as simulate takes them, a signal the design
      has not said and left out, with the interface's messages.
    • Tests: test/ui/sim-wcfg.test.js (Silinx, ISE style and Tasks, in Portuguese: no Save / Load on the toolbar, ⋮
      at its right end, the menu by the mouse and the keyboard, Save and Load as before, the status line unchanged, no
      warnings), test/ui/isim-parts.test.js, test/ui/tasks-sim.test.js and test/ui/toolbars.test.js (the Model field and the line's padding with the ⋮), test/mcp-wcfg.test.js, test/wcfg.test.js (wcfgSignal, placeWcfg on a VHDL
      and a Verilog design).
  • Each drawn board its own SVG ids (#2838): the boards' PCBs got their colour from a gradient with a fixed id
    (b2pcb, nx2pcb, s3epcb, nx3pcb, atpcb, icbpcb); with the Board Emulator and I/O Pin Planning's Board view
    both open the document had each id twice, and the visible board's url(#…) resolved to the hidden tab's gradient,
    so the PCB was painted white. svgEl (web/js/emulator-pcb.js) now gives the ids a drawing defines the drawing's
    own number, in their definitions and their references (scopeIds), and every PCB's fill has a plain fallback
    colour. The editors' drawings (schematic, ASM, FSM, Petri net, truth table) already had ids per instance.
    • Tests: test/emulator-pcb-ids.test.js (every board drawn twice: no id shared, every reference its own
      drawing's, the fallback colour), test/ui/emu-board-ids.test.js (the emulator and the Board view together, tabs
      switched, each closed, on every board, both interfaces, no warnings).
  • A tab's related views, side by side (#2834). A tab's menu, and View for the document in front, list Show … Beside
    for every view related to the document that exists or can be made now (web/js/related-views.js): Boolean algebra →
    its schematic; a state diagram → its ASM chart; an ASM chart → its state diagram; a Petri net → its state machine; a
    drawing → its HDL and the HDL's translation; a generated HDL file, a converted drawing or the table's schematic → the
    drawing; an HDL source → its translation to the other language, whichever the preferred one; a test bench → its unit
    under test; a C / C++ source → its HDL, its ASM chart (sequential), its synthesis report; the views made from it → the
    C; a waveform → the module it shows, or the test bench and its unit under test; the RTL and technology schematics, the
    C model, the power report and the Design Summary → their module. An entry opens the view on the split's other side,
    the document kept in front; a view open already is brought to the front there (a translation made again); one that
    cannot be made says why in the console and leaves the split as it was. The labels replace Show Generated HDL / VHDL /
    Verilog Beside ("Show Schematic Beside", "Show VHDL Beside", "Show Test Bench Beside", …), in Portuguese too. New
    read-only views: a state diagram's ASM chart, an ASM chart's state diagram, a C source's synthesis report as a
    document (web/js/related-docs.js). In the Tasks interface a view shown beside joins its document's task.
    • MCP: related_view gets the read-only views that are no file of the project (an HDL source's translation, a
      state diagram's ASM chart, an ASM chart's or a Petri net's state diagram, a truth table's schematic), nothing
      written; the others are read_file's, c_generated's and c_synthesis_report's.
    • Tests: test/ui/related-views.test.js (per kind of document the entries, each beside, one open already brought to
      the front, a failing conversion in the console, Portuguese; both interfaces), test/ui/split-view.test.js and
      test/ui/translation.test.js (the new labels), test/mcp-views.test.js (each view as the core makes it, nothing
      written, the refusals), test/mcp.test.js and test/mcp-modules.test.js (the tool list).
  • The single-file imports through the import gate (#2830): Import Source / Add Copy of Source (gateFile for each
    file; one refused is not added, a WARNING: line in the console), the Petri net editor's PNML ⇧ (a refused net leaves
    the net as it was), the MCP's write_file (<path>: not written: <reason>). test/import-gate-coverage.test.js
    fails when a module that reads a file from outside (a file input, files dropped, a zip, an upload, the MCP's exchange
    folder, a design link's compressed text) does not import the gate, or is not listed there with the reason it needs
    none.
    • Tests: test/import-gate-coverage.test.js, test/mcp.test.js (write_file: an XXE schematic, a binary source, a
      script in a name, a long elsif chain, a web page), test/ui/projects.test.js (Import Source, both interfaces),
      test/ui/petri.test.js (an XXE PNML).
  • The project imports through the import gate (#2830): the zip upload (/api/projects/import-zip), the .xise with
    its files (/api/projects/import-xise), the browser edition's importZip, importXise and project bundles, and the
    MCP's import_project (a zip, a .xise with its folder, a folder) read their files through core/import-gate.js
    (gateZip, gateEntries); importProjectEntries and projectFromZip take only a gate's result. The gate's refusals
    come first in the import's warnings (the console's and the MCP result's), but those of an ISE project's own outputs,
    which the import leaves out anyway; a refused .xise or silinx.json is the import's error, with its reason. The gate
    keeps a project's documents, scripts and other files as they are (opaque: never shown as a page nor run).
    • Tests: test/server-api.test.js (the zip and .xise routes: an XXE .xise, a traversal and a script in a name,
      every warning asserted), test/mcp-transfer.test.js (a zip bomb, an XXE .xise, a script in a name),
      test/project-zip.test.js, test/project-store.test.js, test/import-gate.test.js, test/ui/projects.test.js (the
      console's refusals, in both interfaces), test/ui/standalone.test.js.
  • The import gate (#2830): core/import-gate.js, the one module every file or project from outside Silinx is to pass
    through before any parser or editor sees it, the same in the browser and on the server: the sizes of a zip, its
    decompressed total, its file count, each file, a source's lines and nesting (an HDL's long elsif chain, #2577),
    checked while the zip is read (a refused entry is never inflated); the names (no .., absolute path, drive letter,
    backslash, control character, hidden folder, reserved name, symbolic link, duplicate but for case); XML with no DTD
    or entity declaration (XXE, billion laughs); each file's type by its name and its content (a web page, a script or a
    program refused; a text that is binary or HTML refused; invalid UTF-8 replaced with a warning). core/zip.js's
    readZip takes the entry count's cap and a check of each entry before it is inflated, and says which entries are
    symbolic links. SECURITY.md describes the threat model and the three layers (the gate, safe output, trust for
    outside projects); the importers move onto the gate in the next pull requests.
    • Tests: test/import-gate.test.js (a test per threat: zip bombs, traversal, symbolic links, XXE and billion
      laughs in every XML format, a script in every file name, oversize and deeply nested sources, types by content;
      the examples and the ISE fixtures accepted with no refusal and no warning), test/zip.test.js.
  • Safe output, checked (#2830): test/safe-output.test.js scans web/js for HTML built from a non-constant string
    (innerHTML / outerHTML assigned, insertAdjacentHTML, document.write, h()'s html:): each one is among the
    reviewed sinks, whose project text goes through an escaping helper (esc, highlightHdl, svgText), or the test
    fails (the reviewed counts only shrink). It checks the Content Security Policy (no inline script but the page's own,
    no remote script, no plugin, no <base>) and that the application's only dynamic code is the compiled simulator's
    (whose generated code holds no name of the design) and the iCE40 nextpnr's own module, so 'unsafe-eval' runs no
    project text. test/ui/safe-output.test.js opens a schematic, an FSM, an ASM chart, a truth table, a Petri net and HDL
    sources with a markup payload in every text field, in the editors, the Design tree and the console: no element is
    made of it. SECURITY.md says how.
    • Tests: test/safe-output.test.js, test/ui/safe-output.test.js (both interfaces).
  • Trust for outside projects (#2830): core/trust.js. Every import marks the project trusted: false in its
    silinx.json (a zip's own silinx.json cannot say otherwise); the local server refuses Xilinx ISE
    (/implement with ISE), /program and /prom for it with a 403 and the reason, the browser edition its WebUSB
    programming, the MCP's run_process with ISE; saving the project's description keeps it untrusted (the server's
    writeProject, the browser stores' saveProject). POST /api/projects/:p/trust (the browser edition's
    trustProject) clears it: the console's line said when the project opens carries a Trust Project button (log()'s
    new action), "Confiar no Projeto" in Portuguese. No MCP tool trusts a project, by design (only the user, in the
    interface). SECURITY.md, GETTING-STARTED and COMECAR say so.
    • Tests: test/server-api.test.js (the refusals, a description written whole, the trust route, a project made
      here), test/project-store.test.js (both browser stores), test/mcp-transfer.test.js (run_process with ISE refused,
      no trust tool), test/ui/projects.test.js (the line and its button, in Portuguese, both interfaces),
      test/ui/standalone.test.js.
  • The power estimate's test on the examples within its time on CI (#2828): since nextpnr is the one engine, the
    test of the estimate on four examples took about 330 s on GitHub's runners against its 300 s. It now places each
    example once (placements, a test's field of the flow's spec, in place of Normal's four), and
    scripts/measure-power.mjs's flow synthesises at Normal, the IDE's default, rather than High: about 70 s
    locally, against 300 s, with the same expected values and no warnings.
    • Tests: test/power.test.js (one placement logged, the estimates within their bounds, no warnings).
  • A design's link, its encoding and the MCP server's tools (#2826, #2829): core/design-link.js puts a whole
    project in a link (#design=: its files as JSON, deflate-raw, base64url, the HDL a drawing makes again left out
    and made again on opening: every example back byte for byte, 5 to 25 KB), reads a link's options (design=,
    open=<https zip>, open=example:<name>, view=try|simulate|emulate, top=, board=) and checks that a design
    has the view it opens; the design's files go through the import gate (core/import-gate.js, #2830); an untrusted link is refused with its reason: too long, a decompression bomb (inflated no
    further than its limit), a file outside the project, a name that is not one, a zip address that is not https or is
    of this computer or a local network. The MCP server's project_link makes a project's link and open_link makes a
    project of a link (never replacing a project that differs; a zip at an address is not fetched).
  • Opening a design from a link, Copy Link… (#2826, #2829): at start-up the address's #design= /
    #open=example: / #open=<https zip> with view=try|simulate|emulate and top= opens the design in a project
    named after it (web/js/link-open.js) and then its view: the schematic's Try it running, the test bench simulated, the
    board emulator; the link cleared from the address. A project of that name with the same files is opened as it is,
    one that differs is replaced only on Replace (Keep Mine keeps it); a zip from another site is fetched only once the
    user agrees, with no credentials nor referrer, at most 16 MB. A broken or hostile link, or a view the design does
    not have, is an error in the console, the workspace as it was. Copy Link… (web/js/link-copy.js) shows the link as it
    is made, its length, and says when it is too long for some slide tools or too large for a link. The local server's
    token is now taken only from an address whose hash is the token alone, never from a link. In both interfaces, in
    Portuguese; GET /api/templates/:name/files (and the browser edition's) gives an example's files.
  • nextpnr's "Max frequency" is no warning on the Spartan-3E (#2820). The nextpnr step logged every Warning: line
    of the kept placement, nextpnr's estimate of each clock among them (WARNING:nextpnr - Max frequency for clock '$iopadmap$clk': … (FAIL at 50.00 MHz)), which can disagree with the timing step (fitted to ISE's trce). It leaves
    it out now, as the Spartan-6's step does; the estimate stays in the step's summary line ("its estimate: … MHz").
    The design corpus's expectations lose the eight such warnings, each beside the timing step's own.
    • Tests: test/nextpnr-fmax-warning.test.js: blinky on the Basys2-250 with a PERIOD of 4 ns, one warning (the
      timing step's, on clk); with no PERIOD, no warning.
  • Tests on the nextpnr engine (#2814). Three test files still built with Silinx's engine, which the application no
    longer chooses (#2137), and failed on main:
    • test/nextpnr-size.test.js compares nextpnr's slices with Silinx's packer's by packing the build's own netlist
      (fpga/pack.js, which stays);
    • test/s6-devices.test.js asserts nextpnr's routing line on the Nexys3;
    • test/compiled-words.test.js places and routes blinky's models through nextpnr, its Post-Place & Route model one
      cell fewer, a constant on routing.
    • Tests: those three.
  • The Spartan-3E with the nextpnr engine: the router's fix of #2800 (#2811). nextpnr-himbaechel-spartan3e 0.20.0
    has the Spartan-6's fix of #2800: a net whose route leaves one of nextpnr's router partitions is routed by its
    whole-device pass instead of being given up. No example hit it on the Spartan-3E, but any design could on some seed.
    The examples route on every seed on the Basys2 and the Nexys2, with no warnings.
  • No problems panel in the schematic editor (#2805): the editor's Check and a refused Try it send their problems
    to the console in every interface (web/js/tasks-check.js initCheckMessages, no longer only in the Silinx one),
    as docs/UI-GUIDELINES.md asks: a header, the errors and warnings, each taking the drawing to its item, and the
    result ("Check Schematic", "Try it"), the Errors tab brought to the front, else the Warnings tab (showProblems,
    #2803). The editor's .se-diags panel is gone (web/js/sch-try.js, web/js/sch-dom.js), with its styles and
    its labels' translations, and so is the refused Try it's message in the editor's hint line.
  • The Petri net's and the state diagram's Check (#2803): a Check button with the shared check icon, after the
    editor's zoom or settings (web/js/petri-editor.js, web/js/fsm-editor.js), hidden in a read-only diagram. Its
    lines go to the Console under Design ("Check Petri Net", "Check State Diagram"), and a click on a line selects
    its item. Its errors and warnings go to the Errors and Warnings tabs (web/js/editor-problems.js). The Petri
    net's validation (core/petri.js) also finds a transition that is not connected and an arc that takes or puts
    more tokens than its place holds. Its Check (core/petri-analysis.js checkPetri) adds the token game's
    transitions that never fire and its first deadlock, with the firing sequence that reaches it, and keeps them in
    the tabs until the net changes. The MCP server's check_drawing (server/mcp-check.js) checks an ASM chart, a
    state diagram or a Petri net the same way.
  • The end of a check brings its tab to the front (#2803): showProblems of web/js/console.js brings the
    Errors tab forward if the check found errors, else the Warnings tab if it found warnings, even when they were
    already listed. Check Syntax (the design, a file, a C source), Check Constraints, the charts' Validate and Check
    and the Schematic's Check in the Silinx interface all end with it. The ASM's Validate no longer brings the
    Console tab in front of its errors. A message written to the console again (a process run again) counts as new,
    so it brings its tab forward too.
  • The Spartan-6 with the nextpnr engine: every seed routes (#2800). nextpnr gave up on some seeds: blinky on the
    Nexys3, seed 1, and clock on the Atlys, seeds 2 and 3, each on a net from an input pad. Place & Route worked only
    because it moved on to the next seed. nextpnr's router now routes those nets on every seed, with no warnings.
  • nextpnr's slices on the Spartan-3E (#2797): nextpnr-spartan v0.18.0 routes the constants that need no LUT as
    Silinx's router ties them (a carry chain's carry in at 0, a pad's output at 0 or 1, a flip-flop's D through BX / BY:
    from a free slice's GND source or the VCC site that reaches the pin), where it made a LUT each, and takes an enable or a
    set / reset active low as the slice's CEINV / SRINV: the flow's netlist turns an inverter on a flip-flop's control pin
    into its IS_<pin>_INVERTED, as Silinx's packer does (#527). A slice that is only a constant 0's source is no longer
    counted as occupied, as ISE's map and Silinx's packer count it. The examples, nextpnr / Silinx: traffic 66 / 66,
    blinky 116 / 116, pong 224 / 225, stopwatch 197 / 197, clock 195 / 196, vga 143 / 144, isqrt 63 / 63, uart 112 / 112,
    micro80 148 / 148 (69, 122, 232, 207, 200, 148, 65, 113 and 154 before); their periods unchanged on average (+0.2 %).
    ISE's DRC passes each, its .bit byte-identical to Silinx's.
    • Tests: test/nextpnr-size.test.js, packages/silinx-fpga/test/nextpnr-engine.test.js, packages/silinx-fpga/test/nextpnr-s3e.test.js, packages/silinx-fpga/test/nextpnr-hard.test.js.
  • MCP run_process: no bare '…' (#2795): its text gave a … line, the separator between the run's errors and
    warnings and the end of its log, also when there were none, the text then beginning with it; it is written only
    after errors or warnings now (processText). Test: test/mcp-modules.test.js.
  • A pad's clock named by its port in the timing report (#2793): the report (fpga/timing-checks.js timingReport)
    named a clock by its net, which Yosys's I/O pad mapping calls $iopadmap$<port>, where its constraint's line and
    ISE's trce name the port; every line that names a clock (its period, its detours, the paths between clocks, its
    hold check) now says the port's name (shownClock), a clock made by logic its net's. The report's data keep the
    nets' names. Tests: test/fpga-timing.test.js, test/open-flow-core.test.js, test/fpga-clock-skew.test.js,
    test/fpga-ff-hold.test.js.
  • The browser edition's cache reads nextpnr's files (#2792). assetFetch, the flow worker's and the Yosys workers'
    fetch that reads and fills Cache Storage, had its own pattern of the versioned assets, written before nextpnr: Yosys,
    the device descriptions and the bit databases, not nextpnr's bundles nor its chip databases
    (vendor/nextpnr-spartan/<version>/gen|chipdb/, vendor/nextpnr-spartan6/…). A project's prefetch put them in the
    cache, but where the service worker did not answer (it reads the same cache), every run fetched them again. One pattern now, VERSIONED_ASSET, for isVersionedAsset
    and assetFetch; assetFetchSource writes it into the Yosys workers' code as a literal.
    • Tests: test/asset-cache.test.js (nextpnr's files of both families, prefetched, read with no request, in the flow
      worker and in a Yosys worker's code; a miss kept; assetFetch and isVersionedAsset agree on every URL, failing on the old code); test/ui/standalone.test.js (#967:
      nextpnr's files counted among the versioned assets Implement must not fetch, and those a run with an empty cache downloads).
  • The Spartan-6 with the nextpnr engine: uart builds on the Atlys (#2791). nextpnr's Spartan-6 package is now
    built from the current bit database, which knows the switches of the Atlys's USB-UART pins (#2539). nextpnr routes
    uart on every seed, on the Nexys3 and the Atlys, with no warnings.
  • The Spartan-6 with the nextpnr engine (#2788): a placement that nextpnr cannot route no longer stops Place &
    Route: the next seed is tried, and the error, when no seed routes, gives nextpnr's reason instead of "undefined".
  • The waveform viewer's status line the shared one (#2785): its line (the simulated time, the cursor, the marker's
    delta, a run's progress) has the documents' status line's class and style (.doc-status, web/css/status-line.css,
    #2763), 20 px high as the others; its test of the theme's wave colour counts the window's pixels in it rather than six
    points a trace may cross.
  • The status line says no messages (#2784): the Schematic editor's status line no longer counts the check's errors
    and warnings; I/O Pin Planning's "pins assigned more than once" is a warning in the console (once, until the pins in
    question change) and its "Saved to …" is gone (the tab's mark says it), its status line keeping "N/M I/Os assigned"
    (docs/UI-GUIDELINES.md).
  • The nextpnr engine's clocks on their dedicated paths (#2770, #2768, #2766), nextpnr-spartan v0.19.0. A DCM's
    CLKIN from its clock pad or from another DCM's buffer, a clock switch's input from its pad and a DCM's outputs into
    their buffers take their dedicated paths (they took general routing, which ISE's DRC passes, with more delay and
    skew); a DCM with five outputs on buffers takes one CLKDUB, not two. A DCM's CLKIN from a pad that does not reach it
    on a dedicated path is said, with the pads that do: DCM u_dcm: its CLKIN comes from the pad B8 through general routing: only B9, A9, A10, C9 reach its CLKIN on a dedicated path. Two IOB registers of one kind on different clocks
    in a LOC'd differential pair are routed: the DIFFM's on its second flip-flop (IFF2, OFF2), as ISE's map puts it.
    • Tests: test/open-flow.test.js (the todos of #2770 and #2766 removed; the clock switch's pad, the five-output
      DCM's one CLKDUB and its warning), test/clocking-flow.test.js (the warning on the XC3S100E again),
      test/ise/open-flow.test.js (the LOC'd pairs, their todo removed), packages/silinx-fpga/test/nextpnr-iob.test.js.
  • The status line, one for every editor (#2763): the Schematic editor's status line becomes a shared module
    (web/js/status-line.js, .doc-status), mounted by the ASM chart editor (its states and every-cycle blocks, its
    encoding), the FSM editor (its states and transitions, Moore or Mealy, its encoding), the Petri net editor (its places,
    transitions and arcs, its tokens) and the truth table (its inputs, outputs and rows), in their read-only views too;
    the counts leave their toolbars (#2725, #2734, #2748). Left out of a print. docs/UI-GUIDELINES.md says what the status
    line holds and what stays in the toolbar.
  • The text editors' status line (#2763): createEditor (web/js/editor.js) gives its editor a status line (the
    shared one of web/js/status-line.js), "Ln 5, Col 3" and the language, put under it by the HDL / C / UCF editor, a C
    source's HDL, a module's C model and the HDL translation; the cursor's place leaves the application's status bar.
  • The viewers' status line (#2763): I/O Pin Planning, the implemented design's FPGA view and the Board Emulator
    mount the shared status line (web/js/status-line.js), their informational texts out of their toolbars (#2725); the
    emulator's time is in it (right below the board in the Tasks interface still, #1022). The status line's style is a file of its own, web/css/status-line.css, which the demo
    pages load too.
  • The design corpus and the synthesis's measurements on the nextpnr engine (#2756). test/designs-corpus/run.mjs
    places and routes each design through the flow's Translate, Map and Place & Route with the nextpnr engine, one
    placement (lib.mjs nextpnrPlaceRoute), not through Silinx's placeRouteBest. Silinx's packer still gives the size
    the runner reads first, to choose a larger device. nextpnr's errors are read for the I/O wrapper ("no free pad left in
    this package") and for the larger device ("no BELs remaining"), and a failure gives nextpnr's reason. The expectations
    are recorded again (the README's results table). 63 of the 67 designs build, 13 smaller or faster. The ITC'99 b14,
    b20, b21 and b22 stop in nextpnr's placer (#2818). Nine are larger or slower than Silinx's engine made them (#2819). Six
    have nextpnr's "Max frequency" warning (#2820). research/synth-quality's abc9.mjs and fsm.mjs place and route
    through the same helper. measure.mjs loses --place, the estimate of Silinx's placer.
    • Tests: test/designs-corpus.test.js (nextpnrPlaceRoute on the gates netlist; a failure as an error).
  • The open flow's check against ISE runs on the nextpnr engine (#2756): npm run check:open-flow and the ISE tests
    build each design as the application does, through nextpnr, not through Silinx's own placer and router.
    • Tests: test/open-flow.test.js, test/blockram-flow.test.js, test/clocking-flow.test.js, test/fpga-place-route.test.js.
  • micro80 on the Spartan-6, checked against ISE (#2755): its register bank's distributed RAM gets its unused data
    inputs tied low, as ISE's map does, so ISE's DRC passes on the open flow's micro80 for the Nexys3 and the Atlys and
    test/ise/examples.test.js's check of it is no longer marked as known to fail.
  • Pointers into arrays in synthesis (#2751): a pointer variable with one target in a difference or a comparison
    with another place in that array (a, a + i, &a[i], m[r] + j, a pointer into it) is a difference or a comparison
    of their offsets; g(a + k) is g(&a[k]), and a moved pointer passed on &a[p_off] (core/c-pointer.js). The callee
    is inlined with its indexes from k (core/c-inline.js); in a combinational module its array is the caller's elements
    from k, a multiplexer of them by k when k is not a constant, written back after the call (core/c-dataflow-expr.js).
    A pointer into a row of an array of arrays has a flat offset, its elements m[k / C][k % C]. A pointer with several
    targets in a difference or a comparison, or a place in an array given to a sub-module, is said; a pointer's own error
    no longer comes with "pointer arithmetic is not possible" too (core/c-check.js). The hint of the pointer errors says so,
    in English and Portuguese.
  • Silinx HLS: a pragma's numeric argument given by a constant (#2750): UNROLL factor=, PIPELINE II=,
    ARRAY_PARTITION and ARRAY_RESHAPE factor= and dim=, STREAM depth=, INTERFACE depth=, and the other pragmas'
    numbers take a global or local const, a static const, C++'s constexpr, an enum's constant, a constant of another
    or of sizeof, as Vitis HLS does (UG1399): the same module as with the number. A name that is not a constant keeps
    its error or warning, which now says "'N' is not a constant".
  • The C simulation's switch: a case of a named constant (#2749): switchStmt in core/c-interp.js folded each
    case's value with no constants in scope, so a case of a global or local const, a static const, a constexpr, an
    enum constant or a sizeof one (#2654) never matched and the default was taken. Each case's value is now
    evaluated by the interpreter, as any expression. test/c-switch-const.test.js compares the C simulation with cc's (enum constants: GCC takes
    no const int as a case label in C) and the named constants' cases with the values written out.
  • The test bench's own names never a port's (#2740): the Test Bench Wizard's test benches (core/testbench.js
    generateTestbench, generateSkeleton) declare their constants, signals, variables and labels with tb_ before
    them when the module has a port of that name (VHDL whatever its case): a port named done (traffic's timer) made
    two signals named done, the clock never ran and Fill Expected from Current Design said the simulation did not
    reach every vector. A test bench without a clash is as it was (the examples' modules: 128 test benches and
    skeletons byte for byte). Tests: test/testbench.test.js, test/mcp-test-bench.test.js.
  • The ASM chart editor's Properties panel (#2739): with nothing selected, no "Selection" heading, help text and box
    legend (web/js/asm-inspector.js: the selection's section hidden); the Machine section a section of its own with its
    heading, not a <details> that folds (web/js/asm-editor.js), in the editor and its read-only views. Tests:
    test/ui/asm-inspector.test.js, test/ui/asm-editor-chrome.test.js (both interfaces).
  • The FSM editor's problems to the console (#2737): the editor gives its validation to onProblems(diags) and no
    longer lists it; drawings.js puts its errors and warnings in the console's Errors and Warnings tabs with the ASM
    editor's editor-problems.js (#2730), in the interface's language (again when it changes), removed when the editor
    closes; a row's click selects the state or the transition (focus(d)). The notes (a state that stays when no
    condition is true) are no longer listed. A read-only view is not validated. Tests: test/ui/fsm.test.js,
    test/ui/new-source-empty.test.js, test/ui/projects.test.js, test/ui/petri-fsm.test.js.
  • The Petri net editor's hint and status lines gone (#2735): as docs/UI-GUIDELINES.md asks and the ASM editor
    did (#2730): its counts are the toolbar's informational text (.asm-info), the zoom level beside the zoom tools,
    the selection's section hidden when nothing is selected, the net's a plain section (not a details).
  • Its problems and messages in the console (#2718): the Problems section is gone; mountPetriEditor says its
    validation with onProblems (when it changes) and its messages with onMessage(text, isError); drawings.js puts
    them in the Errors and Warnings tabs (editor-problems.js, a click selecting the item with ed.select) and the
    Console (an edit refused: a warning; the simulation's: a note), and clears them when the net is closed.
  • An error or a warning written: its tab in front (#2733): the console's shared drawing of its Errors and
    Warnings tabs (web/js/console.js) brings the Errors tab forward for a new error, else the Warnings tab for a new
    warning: once per batch of lines (errors winning), not again for a message already listed, and a warning leaves the
    Errors tab in front while it lists errors, and a tab so brought forward goes back to the Console once it lists
    nothing (cleared, a project opened after the start-up checks' warnings); info never switches it, and the keyboard's focus stays where it is. The
    Silinx interface's folded messages open for a new warning too (for errors only before).
  • The ASM chart editor and its read-only views as the other editors (#2730): no hint text on the canvas and no
    status line (web/js/asm-editor.js); the count of states (and of every-cycle blocks) at the toolbar's right end
    (.asm-info, asm-render.js), the zoom level beside the zoom tools, an "Encoding:" label before the encoding's
    selector. The Problems panel removed: the chart's errors and warnings go to the console's Errors and Warnings tabs
    while it is open, updated as it is edited, a click selecting the box (web/js/editor-problems.js, for the diagram
    editors; part of #2718), and Validate writes them in the Console under Design (Started : "Validate" of …, each line going to its box);
    the editor's notes (a connection dropped on nothing) are Console lines. Portuguese. Tests:
    test/ui/asm-editor-chrome.test.js (both interfaces), test/ui/asm-render.test.js, test/ui/diagrams.test.js,
    test/ui/new-source-empty.test.js, test/ui/c-asm.test.js, test/ui/editor-toolbars.test.js, test/messages.test.js.
  • The FSM editor's layout (#2729): the hint line under the canvas and the status line under the editor are gone,
    in the editor and in a Petri net's state machine (read-only). "N state(s), M transition(s)" is the toolbar's
    informational text at its right end (as #2725); the zoom level follows the zoom tools; "Encoding:" and "Coding
    Style:" label the toolbar's two selectors. Nothing selected, the Properties have no Selection section (its text and
    legend gone); the Machine section is a title, not a fold. The reset's name is drawn on the start-up arrow (fsm-render.js). A text
    typed in place that is not valid goes to the console as a warning (onMessage), no longer to the status line; a
    read-only view's description is disabled too. Portuguese for the new labels. Tests: test/ui/fsm-layout.test.js (both
    interfaces), test/ui/petri-fsm.test.js, test/ui/fsm.test.js.
  • The kind of a state machine derived from its outputs (#2728): fsmKind of core/fsm.js, which normalizeFsm writes
    as the diagram's type; the stored one is not read. A file whose stored kind agrees with its outputs generates the
    same VHDL and Verilog as before (the repository's 23 state machines: the test designs, traffic, the KISS2 corpus,
    the wizard's templates, in every encoding and style). When it contradicts them, the outputs win: a "Mealy" machine
    with no output on a transition is Moore (the same logic, its header saying Moore); a "Moore" machine with an output
    on a transition, an error before, is Mealy and generates its HDL. The Sequence detector template keeps its Moore /
    Mealy choice (it draws two different machines). Tests: test/fsm-kind.test.js (the HDL compiled, elaborated and
    simulated in both languages), test/ui/fsm-side.test.js, test/ui/chart-wizards.test.js.
  • The interface's guidelines: what a toolbar says at its right end (#2725): docs/UI-GUIDELINES.md's toolbar layout
    says a toolbar's informational text (the read-only notice, the sync notice, the counts, I/O Pin Planning's top, device
    and board, the FPGA view's info, the Board Emulator's status) goes at its right end, never between its tools.
  • A toolbar's informational text at its right end (#2725): I/O Pin Planning's "Top: … · Device … · Board: …"
    and its status after the spacer and the ☰ panels' buttons (web/js/pinplanner.js), the FPGA view's info after its
    ☰ Properties, right-justified (web/js/fpgaview.js, fpgaview.css), the truth table's sync notice (truthtable.css);
    the rule in docs/UI-GUIDELINES.md. Test: test/ui/toolbar-info-right.test.js, both interfaces.
  • UI tests in step with main again (#2722): a state's properties in the state diagram editor showed the text
    "null" under its name when it has no comment (since #2696; web/js/fsm-side.js), which the FSM editor's Portuguese
    test caught, now with an assertion of its own; the hierarchy test expects a Petri net's [FSM] badge (#2696); the open
    toolchain's Technology Schematic test, blinky's and pong's cell counts after their counters were narrowed (#2321).
  • The emulator's notes go to the console (#2718): as docs/UI-GUIDELINES.md asks, no message in a document's window.
    mountEmulator says them with onNote(text, kind) (web/js/emulator-side.js, web/js/emulator-serial.js); the
    Board Emulator's document writes them as "Board Emulator: …" lines after its first one (web/js/viewers.js), each run.
    The "Not on the board" box, the engine's and timing's notes, the VGA monitor's warning line and the serial
    terminal's note left the window, with their Portuguese (console lines are not translated).
  • A signal that only feeds a primitive's input is used (#2716): the design checks (core/lint.js) count an input of
    a primitive as read by it, also when its simulation model does not read it: the Clocking Wizard's feedback, a
    DCM_SP's CLKFB with no CLK0 output, a PLL_BASE's CLKFBIN, was said "assigned but never used" (Silinx:646), in VHDL
    and in Verilog. The warnings of the repository's 70 projects (90 views) are unchanged. Tests:
    test/lint-design.test.js, test/mcp-wizards.test.js.
  • I/O Pin Planning: a Ports panel, a bus's bits one by one, a bit taken off its pin (#2715): the Board view's ports
    in a side panel on the left, as the schematic editor's Symbols (its title, folded to a strip and remembered; under
    1024 px over the board, shown by the toolbar's ☰ Ports); a bus
    expanded into its bits, where each one is, each dragged alone (the bus's other bits kept); a port's or a bit's ✕, a
    part's right-click menu and a row of Board Resources' take it off its pin, undoable (a clock's period with it). The
    resources the drawing has no part for (the clocks, VGA, PS/2, the serial port, the Pmods) take a drop on their rows
    in Board Resources; the chips under the board, which repeated that panel, are gone. What a drop places, the pins it
    takes from other ports and its warnings go to the console only, not under the board (core/pin-board.js
    portBit, boardRemove; web/js/pinplanner-ports.js).
  • The pull request template asks for the interface's rules (#2714): a checkbox, a change to the interface keeps
    docs/UI-GUIDELINES.md (its toolbars, read-only views and side panels).
  • The Board Emulator's side panel on a narrow window as the diagram editors' (#2713): under 900 px wide (a tablet,
    the Tasks interface's panel beside it) the panel goes over the board, closed at first, shown and hidden by ☰
    Properties in the emulator's bar (panelToggle, remembered as the editors' Properties; narrowOnly: false, the
    emulator narrow by its own width), instead of below the board with the whole emulator scrolling down; the board's
    area keeps the whole width. Tests: test/ui/emulator-panels.test.js, test/ui/tasks.test.js (the bar's buttons in
    test/ui/emulator-parts.test.js and test/ui/sim-emu.test.js without it).
  • Show Generated VHDL / Verilog Beside (#2711): an HDL source in the other language than the Preferred Language
    has the item in its tab's menu, and View ▸ Show Generated HDL Beside opens its translation too (web/js/split-view.js
    generatedOpener). The translation already open is made again where it is, with the source's unsaved edits
    (web/js/translation.js), rather than closed and opened again.
  • Read-only diagrams without a bar above their toolbar (#2710): a Petri net's state machine and a C source's ASM
    chart show their notice at the right end of the diagram's toolbar (banner of mountFsmEditor and
    mountAsmEditor), as the text editors and the schematic do (#2697), and no longer the document's name, which is
    the tab's; the C source's HDL and the C model's bar without the name too. Test: test/ui/readonly-notice.test.js.
  • The interface's rules for contributors (#2710): docs/UI-GUIDELINES.md, linked from CLAUDE.md and ONBOARDING.md:
    no bar above a document's toolbar, the read-only notice at its right, no Undo / Redo / Print in it, the side panels
    alike (foldable, over the canvas behind a ☰ button of the toolbar on a narrow screen), the hint line, every message
    in the console and never in the document's window, the Design pane's badges.
  • A read-only document's notice at the right end of its toolbar (#2697): one shared style for every
    .gen-banner.read-only of an editor's bar or a schematic's toolbar (margin-left: auto, ellipsized); in the
    schematic, before its ⋮ menu and after the Symbols / Properties buttons of a narrow window.
  • A Petri net's state machine, from its state space (#2696): core/petri-fsm.js petriToFsm(net) explores
    the markings the hardware reaches (the clock step: every enabled transition whose guard is true fires, in
    priority order; clockStepBy, shared with clockStep), a Moore state for each (and for each set of the
    transitions with pulse outputs that fired at the last edge, the pulses being registered), its outputs the net's;
    a transition to each next state, its condition the guards' combinations that lead there, minimized
    (Quine–McCluskey), the combinations no input value gives as don't cares. The machine runs as the net, cycle by
    cycle. The FSM editor has a read-only mode (moved about, zoomed, its states selected, their marking shown, its
    tables and simulation; nothing edited); the Design row's FSM badge, between the net's and its HDL's, opens it.
  • I/O Pin Planning: a Board view, the ports dragged onto the board's parts (#2694): Table and Board in the
    toolbar; the Board view draws the project's board with the Board Emulator's layouts (Basys2, Nexys2, Nexys3, Atlys,
    Spartan-3E Starter Kit, iCEBreaker; another board as rows of LEDs, switches and buttons), the top-level ports beside
    it, dragged with the mouse or a finger. A bus spreads over the row from the part dropped on (moved back to fit
    it); on a digit of the 7-segment display, 7 bits take the segments CA..CG (8 with DP), as many bits as anodes the
    anodes, one bit the digit's anode; the resources the drawing has no part for (clocks, VGA, PS/2, the serial port,
    the Pmods) take a drop on their rows in Board Resources (#2715). A part shows the bit placed on it; a part another port has is taken
    from it; a port dropped again moves. Warned in the console: an output on an input of the
    board (a switch, a button), an input on an output (an LED), a bus wider than the row (the bits left unplaced).
    Undoable; the table and the UCF the same assignments (core/pin-board.js).
  • VHDL ↔ Verilog translation in the Design pane (#2693): the badge of the Preferred Language on the row of an HDL
    module in the other language (web/js/translation.js) opens its file translated by a new translator (core/translate.js,
    translate-to-verilog.js, translate-to-vhdl.js, translate-vhdl-expr.js) working on the parsers' IR: generics and
    parameters, ports, signals, constants, enumerated types (localparams), memories and ROM tables, processes and always
    blocks (clocked, with an asynchronous reset, combinational, a test bench's with waits and delays), if / case / for /
    while, concurrent assignments (when/else, with/select, ?:), instances, generates, functions, the numeric_std
    conversions; VHDL's names in their spelling, reserved words renamed (said in the header), an output read inside the
    module an internal signal, a ?: inside an expression or an expression in a port map an intermediate signal. Records,
    procedures, tasks, casez with don't-care bits and a few more have no translation: the document says which, with the
    line. The comments are not carried over. Every example's design sources translate both ways, and their test benches
    pass with the translations swapped in (test/translate.test.js).
  • A UCF's Check Syntax and Check Constraints: Implementation's lines (#2692): the message centre gives the
    "Check Constraints" header the Implementation source (it was Design's, since #994); run from the Design task of
    the Tasks interface, its lines are shown and the filter turns to Implementation (as a C module's commands, #2262).
  • RTL and Technology Schematic: the "Open ‹module's file›" button removed from the schematic's toolbar, and its
    translation (#2690).
  • The ISE interoperability test expects micro80's warnings (#2688): test/ise/xise-interop.test.js expected no
    warning from ISE's project flow on the micro80 example, which gives the light8080 processor's (the accepted ones of
    #2544). It now expects, on the Basys2-250, the same 41 warnings of XST test/ise/examples.test.js expects (one list,
    in test/ise-micro80-warnings.mjs, shared by both, the top's VHDL path its parameter), and on the Nexys3 the
    Spartan-6's 26 of XST, map and par. Test: test/ise-micro80-warnings.test.js.
  • The Spartan-6's bitstream with two or more block RAMs (#2679): the open flow's bitgen refused micro80 ("2 block
    RAMs: the frame bitgen writes again measured with one"). It now writes the frames ISE's bitgen writes again for any
    number of block RAMs (packages/silinx-fpga), byte-identical with ISE's on micro80 on both boards. micro80 joins
    S6_EXAMPLES (test/examples-lib.mjs): test/examples.test.js builds it on the Nexys3 and the Atlys with no warning,
    test/s6-xdl.test.js takes a SLICEM's distributed RAM settings (its register bank), and test/ise/examples.test.js
    expects ISE's Spartan-6 warnings on it (test/ise-micro80-warnings.mjs, checked on both boards); its open-flow check
    against ISE is known to fail on ISE's DRC (the register bank's RAM32M, #2755).
  • The ISE style interface: the themes System, Light and Dark (#2676): View ▸ Theme in its menus, the theme shown
    following the choice (and the system) as in the Silinx interface, also before the first paint; a dark version of the
    ISE look (web/css/ise-dark.css): the menu bar, the toolbars, the panes and their captions, Processes, the console,
    the dialogs and wizards, the HDL editor, the schematic, FSM, ASM, Petri net and truth table editors, SSim, the board
    emulator, iMPACT and the pin planner; the dark tokens and the documents' dark rules of silinx.css shared by both
    looks. The dark contrast scan runs in the interface the UI tests run in; the theme leaves the interface parity test's
    allow-list (#2450). Tests: test/ui/silinx.test.js, test/ui/dark-contrast.test.js, test/ui/interface-parity.test.js.
  • The ISE style interface, light: the disabled menu items and a selected tree row's twisty readable (#2678): they
    were below a contrast of 3 (2.4 and 1.4). Test: test/ui/dark-contrast.test.js.
  • Silinx HLS's report and options in core/ (#2674): View C Synthesis Report's lines and warnings (cReportLines) and
    the clock, family delays and multiplier blocks a C function is scheduled for (hlsClockOf, hlsFamilyOf,
    hlsMultsOf, of a board and a device) move, as they were, from web/js/c-source.js into core/c-report.js, so that
    the server reports them too (the MCP's C tools, next); c-source.js's hlsClock, hlsFamily and hlsMults call
    them. No change of behaviour (the multiplier blocks never a limit, as before: #2783). Test: test/c-report.test.js.
  • MCP: the conversions (#2674): convert_module converts a module of an HDL file as Convert to Schematic / ASM /
    State Diagram does (core/hdl-to-sch.js laid out with elkjs, core/asm-from-hdl.js, core/fsm.js): the drawing becomes
    the module's base, a file of several design units split; its HDL file stays as it is until the drawing is edited.
    convert_drawing makes a new, unlinked ASM chart of a state diagram, a state diagram of an ASM chart, or the
    Boolean algebra of a state diagram's next-state and output logic. The converters' refusals; a module that is a
    drawing's already is not converted. In server/mcp-convert.js. Test: test/mcp-convert.test.js (stopwatch's and
    traffic's test benches as before the conversion).
  • MCP: the drawings' HDL made again on the server (#2674): write_file of a drawing linked to its HDL file
    makes that file again from it (server/drawing-sync.js, the editors' generators through core/view-hdl.js: the
    repository's drawings give their HDL files byte for byte), and a drawing with errors leaves it as it was, stale:
    Check Syntax and the simulation then stop on it with the interface's error (#303). set_source_properties gives a
    drawing a language of its own, or the project's, and set_design_properties takes preferred_language: the
    drawings and their HDL files in that language, the files in the other one removed (#220, #1477, #2254). The
    test benches of gates, blinky and traffic pass in both languages. Design Properties' warning to check the pins
    comes after "Device set to", as in the interface. Tests: test/drawing-sync.test.js, test/mcp-drawings.test.js.
  • MCP: forces (#2674): simulate takes forces: a value, or a clock (its leading and trailing values, period,
    duty cycle), from a start until a cancel, over what drives the signal, as SSim's Force Constant / Force Clock
    (ssim force add) through core/ssim-forces.js; a later force of a signal cuts its earlier one, a released signal
    takes what the design gives it again. Test: test/mcp-forces.test.js (VHDL and Verilog).
  • MCP: Silinx HLS (#2674): c_simulation runs a C source's main() (core/c-interp.js: its output, what main()
    returned, its calls of the top function), c_cosimulation replays them on the HDL made from it in VHDL or Verilog
    (core/c-cosim.js: PASS or FAIL with each mismatch, the latency), c_synthesis_report says what it is made into
    (core/c-report.js) and c_generated gives the HDL or the ASM chart made of it; scheduled for the board's clock,
    the device's delays and the project's RTL blackboxes, as in the interface, with its messages. In
    server/mcp-hls.js. Test: test/mcp-hls.test.js (the test designs fir, gcd, sort, popcount and stream_sum).
  • MCP: a project's import and export (#2674): export_project writes a project as a .zip on the disk, as
    Export Xilinx ISE Project (its .xise, the schematics as ISE schematics, a C source as its HDL) or Export Silinx
    ISE Project does, never over an existing file; import_project makes a new project from a .zip, a .xise with
    its folder or a project folder, as Import Silinx / Xilinx ISE Project does: the paths that leave the project
    and the hidden files dropped, with the import's warnings and the files the project lists but were not found.
    Where the interface has the user choose the file, the assistant names it: both tools take only the paths of one
    exchange folder (the user's Downloads, or SILINX_MCP_FOLDER), its symbolic links that leave it refused.
    The export.zip and import-zip routes' code moves, as it was, into server/project-zip.js, which both call.
    Tests: test/mcp-transfer.test.js, test/project-zip.test.js.
  • MCP: New Source (#2674): new_source makes each type of the New Source wizard from its own templates
    (web/js/templates.js, c-templates.js) and the drawings' models, test benches with the wizard's unit under test
    (core/uut-info.js), in the project's Preferred Language or the one given, with the wizard's checks (a file name,
    a reserved word, a module or a drawing's HDL file already there); a file that exists is not replaced, and a
    constraints file is the project's when it has none (a PCF for an iCE40). In server/mcp-new-source.js. Test:
    test/mcp-new-source.test.js (the module and its test bench compiled, elaborated and simulated in VHDL and in
    Verilog; the drawings' Check Syntax).
  • MCP: the project's settings and sources (#2674): nine new tools of the MCP server. list_devices lists the
    boards (their variants) and the devices by family; set_design_properties sets a board or a device and the
    synthesis tool as Design Properties does, the new device checked against the design (its warnings), the
    constraints file of the device's format (a PCF for an iCE40) and, with update_constraints, the new board's
    pins written into it; set_process_properties sets the Optimization (Fast, Area, Speed), the FPGA start-up
    clock and the place and route engine. unset_top, remove_from_project (a drawing with its HDL file),
    add_to_project, rename_file (a file or a folder: the drawings' links follow), delete_file (a file or a
    folder) and delete_project (to the workspace's .trash) do what the Design pane's menus do; a top module the
    project no longer has becomes the one root left of its view, else none, with the interface's messages. In
    server/mcp-settings.js and server/mcp-sources.js. Tests: test/mcp-settings.test.js, test/mcp-sources.test.js.
  • MCP: a module renamed with its references (#2674): rename_module renames a module of the project's HDL
    files as the interface's Rename does (server/rename-module.js, web/js/rename.js's renameSource on the server):
    its declaration and instances in every source (core/rename.js), the schematics' symbols of it, the drawing it is
    made from (named after it, renamed with it; its HDL made again), the project's top, test bench and stimuli; its
    file follows the new name when it is named after the module, or goes where it is told, or stays. The dialog's
    refusals: a bad name, a reserved word of VHDL, Verilog or SystemVerilog, a module or a file that exists; the files
    are moved all or nothing. Tests: test/rename-module.test.js, test/mcp-sources.test.js.
  • MCP: the simulation models (#2674): run_process makes each model (post_synthesis_model, post_translate_model,
    post_map_model, post_par_model: the flow up to its stage, then the model) and simulate takes model: the test
    bench against that netlist as Simulate Post-… Model does (compiled after the sources, its nets single bits, the
    post-place & route model with its SDF's delays), generate making it first; a model older than the sources is
    said; vcd writes the waves into the exchange folder (a new file). Test: test/mcp-simulate-models.test.js.
  • The MCP's simulation reads the project's data files (#2778): simulate gave the simulator only the HDL, and a
    test bench's $readmemh of a .mem said it was not found; it now gets the data files the interface gives it.
  • MCP: the Test Bench Wizard and the UCF Wizard (#2674): test_bench_wizard writes a test bench of a module
    through core/testbench.js as the wizard does: its clock and reset guessed or given, the vectors generated or
    given, their expected outputs given or taken from a simulation of the design (Fill Expected from Current
    Design), or a skeleton; it reports each mismatch and TEST PASSED / TEST FAILED. constraints_wizard writes a UCF
    of a module's ports on the board's pins (assigned by their names and directions, the clock's period), pins
    given overriding them, as the UCF Wizard does. In server/mcp-test-bench.js and server/mcp-wizards.js. Test:
    test/mcp-test-bench.test.js. A module with a port named done gets a test bench that never ends: #2740.
  • MCP: the Clocking Wizard and the Block Memory wizard (#2674): clock_manager makes a DCM_SP or a PLL_BASE
    module for the frequencies asked (core/clocking.js: each output at the nearest frequency it can, said when it is
    not the one asked), block_memory a RAM or a ROM of the device's block RAMs (core/blockram.js: one port or two,
    its write mode, the primitives or inferable HDL, its initial contents from a .mem or .coe file), in VHDL or
    Verilog; edit makes a module of theirs again from the settings in its header with those given changed, as
    editing it reopens the wizard. The wizards' checks and messages (the device's family and its DCMs, PLLs and block
    RAMs, a memory larger than the device, the ports' names). In server/mcp-wizards.js. Test:
    test/mcp-wizards.test.js (the modules compiled in both languages and simulated: the ROMs read back their .mem
    file, the DCMs lock and make their 100 MHz). A wrong warning of their Check Syntax found on the way: #2716.
  • A project rule: what the interface does, the MCP server does too (#2674): CLAUDE.md's Programming section
    says that every action of the interface is a tool of the MCP server, doing it through the same functions, and
    that a new or changed feature ships with its MCP tool and that tool's tests in the same pull request.
  • SSim's times, values and forces in core/ (#2674): parseTime, parseValue and UNITS move, as they were, from
    web/js/isim.js into core/ssim-values.js, and the signals' and instances' paths and the forces' pseudo-process
    (findSignalIn, findInstIn, injectForceIn) from web/js/isim-forces.js into core/ssim-forces.js, so that the
    server forces signals as SSim does (the MCP's forces, next); isim.js still exports the three. No change of
    behaviour. Test: test/ssim-forces.test.js.
  • A type's text and a test bench's unit under test in core/ (#2674): typeText (with its range and short
    expression helpers) moves, as it was, from web/js/editor.js into core/type-text.js, and the New Source
    wizard's uutInfo from web/js/new-source.js into core/uut-info.js, so that the server makes a test bench as
    the wizard does (the MCP's New Source, next); editor.js still exports typeText. No change of behaviour. Test:
    test/type-text.test.js.
  • Board Emulator laid out as the diagram editors (#2666): its side column is a side panel at the board's right
    (web/js/emulator-side.js), folded as the ASM, FSM and Petri net editors' Properties (side-fold.js) and remembered,
    below the board when the emulator is under 900 px wide (a container query); the board's area scrolls on its own.
    The toolbar's menu (toolbar-menu.js) holds the engine, what it does when hidden and the modules shown
    (web/js/emulator-toolbar.js, emulator-modules.js). The print view keeps the Watch; the Tasks interface fits the
    board to the room left of the panel (web/js/tasks-emu.js). Test: test/ui/emulator-panels.test.js.
  • Pointer arithmetic in the C simulation (#2655): an array, a pointer into one or a string literal plus or minus an
    integer (a + k, k + a, p - k, &a[k], one past the end too) is a pointer to that element; p++, p += n,
    the difference of two pointers into one array (a ptrdiff_t), their comparisons, == / != of any two and of a
    pointer and NULL; such pointers given to functions, to strlen, strcat, memcpy and printf's %s; into a row of an
    array of arrays (m[1] + 2) and into an array of strings (const char **n = names + 1). A pointer moved outside its
    array, read or written one past its end, or pointers into different arrays subtracted or compared, stop the run
    with their place (core/c-interp-expr.js). In synthesis int *p = k + a is taken as a + k (core/c-pointer.js);
    p - a, p < q, g(a + 1) and m[1] + 2 are still refused there, said (#2751).
  • Silinx HLS: the constants of sizeof (#2654): a global constant's initializer has its sizeof of an array, a
    row or a pointer made sizeof of its type (core/c-flatten.js), as the functions' bodies had (#2641); synthesis's
    constants for a case fold sizeof; an enum constant of sizeof becomes a constant declaration
    (core/c-struct.js) instead of the error "its value must be a constant". static const, const unsigned, C++
    constexpr, a constant of another, a struct's and a row's sizeof: no message, a literal's trip counts and
    latency, co-simulated against cc.
  • The workflows that publish hold their writing token in a job of their own (#2644): the development build, the
    release and silinx-fpga's npm release build in a job that only reads, and a job that runs no code of the repository
    or of its dependencies publishes the artifact (the tag dev moved by the API, the attestation, the release, npm publish of the packed tarball); the Yosys update's report job runs the base's scripts only, and checks the pull
    request out no more. A test checks that no job whose token writes installs a dependency.
  • The Vitis HLS image's base pinned by its digest (#2645), as the ISE image's, and updated by Dependabot.
  • The nextpnr engine leaves a block RAM's undriven data input unconnected (#2642), as Silinx's engine does: ISE's
    DRC had warned of a dangling output pin on a 16-bit RAM inferred onto a block RAM.
    • Tests: packages/silinx-fpga/test/nextpnr-hard.test.js, packages/silinx-fpga/test/nextpnr-engine.test.js.
  • Spartan-6: a carry chain's carry in and its first data input (#2638): both used to need the slice's bypass input
    AX, and the implementation stopped with "its pin AX carries … and …". The first data input now goes through the LUT's
    O5, as ISE's map does it (packages/silinx-fpga/changes/2638-s6-carry-in-di0.md). Tests: test/s6-pack.test.js,
    test/s6-xdl.test.js.
  • Tests: the #1220 Unset as Top Module UI test waits for the hierarchy's top module redraw (#2630): in the ISE
    style interface, setting the top module redraws the hierarchy tree with the new top first; waiting for the row to
    become top before clicking avoids a race where the click missed 'b'.
  • nextpnr-spartan v0.17.0 (#2626, #2628, #2656): nextpnr's router no longer loops on a carry multiplexer whose DI
    is its own select or on a 36-bit block RAM beside its tile's multiplier, a slice's two memories share one enable
    constant, and a DCM's clock input from the other edge is routed with no warning.
    • Tests: packages/silinx-fpga/test/nextpnr-engine.test.js.
  • A test's selectors escaped in full (#2623). test/silinx.test.js made a RegExp of each board emulator selector,
    escaping only . and * (CodeQL's js/incomplete-sanitization: a backslash was left as it was). Every RegExp character
    is escaped now, the backslash too, and the test checks it on selectors that have them. Tests: test/silinx.test.js.
  • Interface parity, a test (#2450): test/ui/interface-parity.test.js collects, in both interfaces, every action a
    user can reach with a project open and each kind of editor open (the menus, the toolbars and their "more" menus, the
    panes' and the editors' context menus, the Processes, the command search) and fails on any difference that is not
    in its allow-list, each entry with its reason; every command of the command search runs in both. The gap it found:
    Add to Project… (#1214) had no way in the ISE style interface, now in the Project menu; iMPACT's folded settings
    button in the Silinx interface says whether they are shown (aria-expanded).
  • The Spartan-3E's nextpnr engine at 1.0.0 (#2438). nextpnr-himbaechel-spartan3e 1.0.0 is 0.20.0's place and route
    with a stable interface: a change Silinx would need outside it is a 2.0.0. The designs place and route as with 0.20.0.
  • The RTL blackbox through synthesis (#2341): the blackbox (#1903) had been checked by co-simulation only. A project
    with one did not build: its C model (the description's c_files, the C simulation's) was made a module of its own,
    refused ("'adder' is the top function", #2854). blackboxModels (core/c-blackbox.js) lists the C models the descriptions
    name, and the build leaves them out, on the server (viewSources, the Silinx toolchain and ISE) and in the browser
    edition. An HLS module calling a module in the other language (a VHDL machine and a Verilog module, a Verilog machine
    and a VHDL module, test/fixtures/hls-blackbox) builds with the Silinx toolchain on every device of the examples,
    and with ISE (XST), with no warnings.
    • The output's capture (#2862): an output with a valid was captured, for a valid before ap_done, in registers
      ap_done cleared first: with the usual module, its valid with ap_done, the capture never happened, and XST
      trimmed it with a warning per bit (Xst:1710, Xst:1895). It is now taken on every valid and forgotten as the next
      call starts, that start's cycle reading the valid alone.
    • The corpus against Vitis HLS: misc-rtl-as-blackbox run again through Vitis HLS 2025.2: its C simulation,
      synthesis and co-simulation pass, Silinx's ports the same as Vitis's, a latency of 5 cycles to Vitis's 4
      (test/hls-corpus/vitis.json, the README's table). scripts/hls-vitis.mjs synthesized the design without its
      blackbox descriptions to compare the ports: it now takes the options the corpus runs with (optionsOf,
      test/hls-corpus/lib.mjs, #2855).
    • Tests: test/hls-blackbox-flow.test.js (the C model not a module; the C simulation and co-simulation; the
      Silinx toolchain on the six boards, no warnings), test/ise/hls-blackbox.test.js (ISE on the six boards, no
      warnings; opt-in), test/hls-vitis.test.js (a blackbox design's ports compared), test/c-blackbox.test.js (the capture; calls
      of a combinational module whose valid is 0 on some, none reading the last capture), test/ui/c-source.test.js (the
      C model no row of the Design pane, no error).
  • The examples on the Spartan-6 with ISE: no warning (#2321): XST on the Spartan-6 trimmed the counter bits that
    are always 0 (Xst:1293, Xst:1896) and map warned of the stopwatch's 9K block RAM (PhysDesignRules:2410). The
    button debouncers of clock and pong count in 20 bits (24 before), pong's step and beat counters in 23 (24), blinky's
    ASM chart's ticks register in 7 bits (8; speed_ctrl.vhd generated again), and the stopwatch's lap memory holds
    1023 times in 1024 words (512 x 14 took a RAMB8BWER, 1024 x 14 a whole RAMB16BWER; on the Spartan-3E one RAMB16
    either way). The open toolchain's netlists of clock, pong and blinky are smaller; the stopwatch's 2 LUTs larger on
    the Spartan-3E (its Speed mapping 28), 3 on the Spartan-6.
    test/ise/examples.test.js builds every example with ISE on the Nexys3 and the Atlys too, with no warning, and
    test/ise/xise-interop.test.js expects none on the Nexys3; test/synth-brams-s6.test.js keeps a 512 x 14 memory of
    its own for the open synthesis's 9K rule (#2130).
  • nextpnr-himbaechel-spartan6 in the browser (#2288): the flow's worker gives the Spartan-6's nextpnr engine its
    io.nextpnr.spartan6, from the package's bundle and the device's chip database (about 5 MB, fetched on its first use,
    then kept by the browser's cache). The local server serves the package at /vendor/nextpnr-spartan6, and the standalone
    build publishes it with the site's assets (assets/vendor/nextpnr-spartan6/<its version>/gen and chipdb, each file's
    SHA-256 in the page). When a project opens, the browser fetches nextpnr ahead only for a Spartan-6 project that chooses
    the nextpnr engine (its auto is still the Silinx engine's on the Spartan-6, #2131).
    • Tests: test/ui/standalone.test.js (traffic on the Nexys3 by nextpnr in the browser edition), test/web-open-flow-browser.test.js, test/asset-cache.test.js, test/family-profile.test.js.
  • The Post-Place & Route model's design by nextpnr (#2137). When the test bench gives the top other generics, the
    design synthesized with them is placed and routed for its model by the family's back end, nextpnr, as Implement does
    (placeRouteNetlist, a report and a log of its own), no longer by Silinx's own placer and router. The console says
    "Placing and routing the design synthesized with the test bench's generics (nextpnr), for its model."; a failure names
    nextpnr's reason.
    • Tests: test/sim-model-par.test.js (the examples' test benches on their post-place & route models, through
      nextpnr; blinky's with its test bench's generics).
  • nextpnr, the application's one engine (#2137). engineOf gives nextpnr for every project. A project's
    impl.engine (silinx, auto, nextpnr: #2136's choice) is no longer read, and Process Properties' OK drops it. The
    auto fallback to the Silinx engine is gone: when nextpnr cannot start, Place & Route fails with "nextpnr's chip
    database of could not be downloaded (…)" and a hint to implement again online. The nextpnr steps' hints no longer
    say "Choose the Silinx engine". The Spartan-6 places and routes with nextpnr too (its speed gap, #2798, accepted), and
    the browser fetches its nextpnr ahead when a project opens. MCP: set_process_properties loses pr_engine. The command
    line's --engine default is nextpnr and auto is refused; --engine silinx stays until Silinx's placer and router are
    removed. Register replication (#4), the Silinx engine's only, leaves the application with it.
    • Tests: test/ui/open-toolchain.test.js (no engine in Process Properties, in Portuguese neither; a project saved
      with the Silinx engine implements with nextpnr), test/ui/standalone.test.js, test/mcp-settings.test.js,
      test/open-toolchain.test.js, test/silinx-fpga-cli.test.js, test/web-open-flow-browser.test.js,
      test/family-profile.test.js, packages/silinx-fpga/test/nextpnr-engine.test.js.
  • Silinx's Spartan-3E placer and router removed (#2137). nextpnr has been the application's one place and route
    engine since #2810; its code, its tests and the two research scripts that placed and routed with it
    (research/s3e-place/flow.mjs, research/s3e-route/route.mjs) are gone. Register replication (#4) went with it. The
    parts other modules use stay, moved as they were: the device's sites and banks (fpga/device-sites.js), the placers'
    common parts (fpga/place-common.js, the Spartan-6's placer's), a routed design's check (fpga/route-check.js), and
    the clock network's reach and ISE's clock pad per DCM (fpga/clock-reach.js, which the device tests and the research's
    generators of ISE designs read). Silinx's packer stays (#2529, #2530), and so does the Spartan-6's own engine. The bank
    VCCO tests (#128) run on nextpnr: two LOC'd I/Os of one bank with different VCCO are accepted by nextpnr (#2813, a todo),
    and its message for an I/O with no pad of its VCCO names the cell (#2816).
    • Tests: packages/silinx-fpga/test/engine-shared-parts.test.js, packages/silinx-fpga/test/fpga-hard.test.js,
      packages/silinx-fpga/test/silinx-fpga-api.test.js, test/designs-corpus-fixes.test.js, test/open-flow-core.test.js,
      test/iob-pair-clocks.test.js, test/iob-ddr.test.js, test/fpga-timing.test.js, test/silinx-fpga-cli.test.js.

Silinx ISE (Integrated Synthesis Environment) is an ISE-style FPGA design environment that runs in the browser, made for teaching digital systems: VHDL/Verilog projects; schematics with live simulation; FSM and ASM state-machine editors, truth tables and Karnaugh maps, all kept in sync with their HDL; module and test bench wizards; behavioural and netlist simulation with waveforms; a board emulator; synthesis, place & route and bitstreams with its own toolchain (the Basys2 and Nexys2 boards) or with Xilinx ISE 14.7; and board programming, also from the browser.

Downloads

File What it is
Silinx-ISE.html Single-file edition, the same app as the site (https://pmnmalo.github.io/Silinx-ISE/). Download it and open it in a browser: design, simulation, synthesis, place & route and bitstream (the Silinx toolchain, Basys2 / Nexys2; its tools come from the site the first time). Projects are kept in the browser (File ▸ Download Project Bundle to back them up) or, in Chrome / Edge, in a folder on the disk. Programming the board's FPGA and Platform Flash over USB works from the site's address (Chrome / Edge, Chrome on Android).
silinx-ise-<version>.zip Silinx run locally, with its dependencies: an alternative to the site (offline synthesis, projects as folders, the installed programmers, Xilinx ISE 14.7 modes). Needs Node.js 18 or newer. Unzip, then double-click Start Silinx-ISE.bat (Windows), Start Silinx-ISE.command (macOS) or start-silinx-ise.sh (Linux) in the silinx-ise folder; Silinx opens at http://127.0.0.1:8642. Projects are stored in ~/Silinx-projects.
silinx-ise-docker-kit-<version>.zip Kit to build your own private Docker image of Xilinx ISE 14.7 (Windows, macOS Intel/Apple Silicon, Linux) for synthesis, place & route and bitstreams. ISE and its licence are not included and must not be redistributed: download them with your own AMD account. See ise/README.md.

The source code is attached by GitHub below (Source code zip / tar.gz).

Silinx ISE is free software (GNU AGPL-3.0, Copyright 2026 Pedro Maló; see LICENSE, NOTICE and THIRD-PARTY-NOTICES.md). It is an independent project, not affiliated with or endorsed by AMD/Xilinx. Xilinx and ISE are trademarks of AMD.

Checking a download

The files' SHA-256 checksums (the file SHA256SUMS):

dc65d8f51e8fb4a88ae8f1534384929c4bb11f39f298733756a596568fe78e5c  Silinx-ISE.html
178e1fc55a61068060cddc4f9e075e697b66efd883027f64208d03fc117e0f2a  silinx-ise-15.22.0.zip
cd99c48df8b2f500384a69ec82216e9a59f427b9b26eee8b48e81687b63e0571  silinx-ise-docker-kit-15.22.0.zip

In the folder of the downloads: sha256sum -c SHA256SUMS --ignore-missing (Linux), shasum -a 256 -c SHA256SUMS --ignore-missing (macOS), Get-FileHash <file> (Windows PowerShell, compared by eye). Each file also has a build provenance attestation (it was built by this repository's Release workflow from this tag): gh attestation verify <file> --repo pmnmalo/Silinx-ISE.