Skip to content

Releases

CMIS Module Manager v2.213.0

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 05 Oct 06:13

CMIS 5.4 standards audit and optimization

  • SNR reads skip reserved bytes and unsupported measurement sides. Dual-side data READs drop from 32 to 16 per Bank; each U16 still uses its own two-byte READ for coherent monitoring values.
  • Optical power readings remain visible when alarm thresholds are unavailable, without fixed-reference alarm colouring. Failed threshold reads after an Application change discard obsolete limits and retry on later monitoring refreshes.
  • Raw reads retain Network Path and VDM latched events, including when a later read fails. The event API records Bank, Page, address, bits and timestamps; full VDM/Network Path decoding is not added.
  • Updated operator manual and companion Skill. Self-contained 32-bit Windows executable; use a browser to operate the local Flask application.

规范检查与优化

  • SNR 跳过保留区和未支持的测量侧;双侧每 Bank 数据读取由 32 次减至 16 次,并保持逐读数一致性。
  • 功率阈值未知时保留读数、不借用固定值判告警;新 Application 的阈值读取失败后清除旧判据并重试。
  • 原始读取 Network Path / VDM 锁存标志时保全已取得的事件,后续读取失败不会丢失。当前仅保存原始事件,未新增完整解码面板。
  • 同步操作手册与配套 Skill;软件按 MIT License 开源。

Validation: 3008 main regression cases, 30 audit/optimization cases on 32-bit Python, and 254 HTTP requests across all 12 Mock profiles through the rebuilt customer executable. The new table citation was verified against CMIS 5.4 and its citation checks rerun. Browser checks cover multi-Bank SNR/monitoring and absent-monitor rendering. Real hardware was not connected.

Assets: the application ZIP is flat (executable at ZIP root). The Skill ZIP is distributed separately.

v2.212.0 — Customer build and documentation maintenance

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 05 Oct 02:58

v2.212.0 is a maintenance release with a rebuilt 32-bit Windows customer executable and synchronized documentation.

  • Update the operator manual version and date to match this release.
  • Correct the hardware-backend count and document the audit regression test command.
  • Include the matching CMIS skill package alongside the flat application ZIP.

Management features and register handling are unchanged from v2.211.0.

Validation: a 3,008-test source sweep, with the version consistency check rerun successfully after the version bump; 21 audit regressions; 22 regressions on 32-bit Python; 254 HTTP requests to the actual customer EXE across all twelve Mock modules, including thirteen correct flat-memory refusals. Both ZIPs pass integrity and payload checks.

Actual browser automation could not access the local service. No real hardware or PHY conformance validation was performed.

v2.211.0 — CMIS audit fixes and Mock diagnostics

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 01 Oct 15:24

v2.211.0 corrects the confirmed CMIS and Mock audit findings and strict numeric input handling.

  • Keep the complete Data Path deinitialized unless ConfigSuccess is confirmed.
  • Retain COR events immediately, including when later auxiliary or chunk reads fail.
  • Verify each Bank/Page mapping, reject short transfers, failed discovery and future CMIS major versions.
  • Share reset hold-off and rediscovery across control/raw writes and autonomous restarts; use a precise Windows retry clock.
  • Honor advertised CDB trigger formats and return a nonzero exit code for failing tests.
  • Correct ZR Applications to 800ZR-A/B, wavelength advertisements and SR8 range; identify the sixteen-lane fixture as two 800G Applications.
  • Select synthetic bitstream rates by Application, direction and Checker position; count complete absolute gate intervals independent of polling.
  • Explain encoder/decoder positions, synthetic measurement planes and reset freezing; document twelve mocks and the dj D3.1 scope.
  • Parse numeric inputs strictly and reject truncated, trailing-character and out-of-range values.

Validation: 3,008-test full sweep (no errors or skips; the only failure was a stale README test count, corrected and rechecked successfully), plus 30 regressions on the 32-bit Python used for the customer EXE. Customer EXE browser verification covers all twelve mocks and the repaired control/diagnostic workflows.

Mock results are synthetic management fixtures. IEEE P802.3dj references are limited to D3.1; no real-hardware or PHY compliance certification is claimed.

v2.210.0

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 28 Sep 00:02

The FAQ answers and the lane breakdown note say what the tool does

Several answers in chapter 13 had fallen behind the tool:

  • Q3 said the Backend menu lists four mock modules and no USB adapters. It actually always lists all twelve demo modules and all six adapters. Any adapter the tool cannot find is greyed out and marked "(unavailable)".
  • Q5 said "-inf dBm". The page prints "−∞ dBm".
  • Q7 said only mock_coherent_zr is tunable and that tunability comes from the Media Interface Technology. mock_zr16 is tunable too. The tool decides tunability from 01h:155.6, and the Laser Tuning card says "Non-tunable module".

In Module Info, the note on the Host Lanes row said the Applications in the breakdown are alternatives and "not added". On a 2×400G module the two run side by side, and the eight beside them is exactly their sum. The note now describes both cases.

A test checks each answer against the backend registry, the tunable advertisement of every demo module, and the page's own strings.

This is a documentation fix only; the tool behaves as before.

2973 end-to-end tests. 32-bit build, as CH341 requires.

v2.209.0

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 27 Sep 23:40

The raw write chapter names every refusal instead of counting five

Chapter 11 of the manual said a raw register write is refused in "only five" cases and listed five. v2.204.0 added more refusals and the list did not follow:

  • a write that names no page or address;
  • a data byte outside 00h–FFh;
  • a page, address or bank that is not a whole number.

Two older refusals were also missing from that list: a write on Page 9Fh that runs over byte 129, and a Bank that names nothing. They were only mentioned elsewhere in the manual.

The chapter no longer gives a number. It lists each refusal together with the message the tool shows for it, and a test checks that every refusal message in the code is named in the chapter.

The section on which Data Paths an Apply restarts said there are two ways a module silently drops an Apply. It now also names the third: an Apply that arrives while the last command is still ConfigInProgress (6.2.4.2), which the tool refuses with a 409.

This is a documentation fix only; the tool behaves as before.

2969 end-to-end tests. 32-bit build, as CH341 requires.

v2.208.0

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 27 Sep 23:05

The manual's table of traffic-disrupting buttons lists every such button

The appendix table of buttons that affect traffic was out of date in three ways:

  • The DataPath Apply row still said it "restarts all 8 lanes". The chapter was corrected in v2.207.0, but this row was missed.
  • It said only Reset Module asks for confirmation. The media lane Commit does too.
  • It left out Output Controls Apply (forced squelch, Rx output disable), the media lane Commit, PRBS Apply (a generator replaces the traffic) and Laser Tuning Apply.

The table now lists all of them, the DataPath row says what is restarted, and the confirmation sentence names both dialogs. This is a documentation fix only; the tool behaves as before.

2968 end-to-end tests. 32-bit build, as CH341 requires.

v2.207.0

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 27 Sep 22:43

A polarity or Tx disable change no longer restarts every Data Path

CMIS Table 8-77 lists 10h:129-142 as Lane-Specific Control, "independent of the Data Path State machine or control sets". Polarity and Tx disable take effect when written and need no Apply. The DataPath panel writes them through its Apply button, which always asks for an Apply. The server treated "no staged configuration changed" as a request to re-commission. So flipping one lane's polarity restarted every Data Path on the module; on a 2×400G module both 400G ports dropped. The manual has claimed otherwise since v2.10.1.

Now

  • A request whose only change is a lane control (polarity or Tx disable) is written without an ApplyDPInit trigger. It answers applied_lanes: [], and the page says no Data Path was restarted.
  • Pressing Apply with nothing changed still re-commissions every Data Path.
  • A change to the staged configuration applies only the Data Paths it touches.
  • DP Deinit changes keep the order introduced in v2.200.0.
  • The manual's three passages saying Apply writes 0xFF and restarts every lane are rewritten.

2967 end-to-end tests. 32-bit build, as CH341 requires.

v2.206.0

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 27 Sep 21:39

The DataPath table renders when a lane is kept in another NAD Block

v2.202.0 added a dropdown entry that keeps a lane staged in another NAD Block. That entry used two names, bankOf and appName, which were declared further down the same function. On the first such lane the browser threw "Cannot access 'bankOf' before initialization" and the whole DataPath table showed nothing. It happened on exactly the modules the entry was written for. The earlier tests read app.js as text, which cannot show declaration order. Walking the tab in the browser on the 24-lane demo with Block 1 staged is what exposed it.

Now

  • Both names are declared before the dropdown is built. Lane 1 shows "AN 16 (NAD Block 1, AppSel 1) — kept" with its media lanes and "running App 1".
  • The page's loaders are run for real in the tests. app.js is loaded into Node under a catch-all DOM, and every GET is answered with the server's own reply for that path.
  • Each loader runs against every demo module, against a lane kept in another NAD Block, and against a Network Path Application. A loader that throws is a panel that does not render. A self-test checks that the harness does catch one.

2955 end-to-end tests. 32-bit build, as CH341 requires.

v2.205.0

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 27 Sep 21:10

The laser and connect fields are checked like the others

The sweep behind v2.204.0 tested /api/module/laser on a module that is not tunable. There every request is refused for that reason alone, so the sweep said nothing about the fields themselves. On the tunable demo module, the endpoint behaved the way the others had before v2.204.0:

  • grid_code, channel, fine_offset_ghz and target_power_dbm set to a string or null gave a 500.
  • channel: 0.5 became channel 0.
  • true became 1 GHz of fine offset or 1 dBm of target power, and was written.
  • grid_code: 21 went down as 5 (100 GHz) through & 0x0F.
  • fine_tuning_enabled: "no" turned fine tuning on.

Now

  • Each lane entry is checked before anything is read or written: lane and channel are whole numbers, and grid_code is 0–15 (12h:128 bits 7–4).
  • Offset and power are numbers other than true. NaN and infinity reach the advertised range check, which refuses them.
  • fine_tuning_enabled takes true or false only.
  • /api/connect used to hand true, 1.5, −1 and 256 to the adapter as a bus number or I2C address. The address must now be a 7-bit address from 0 to 127 (the module answers at 80, 50h), and the bus 0 or more. Hex strings such as "0x50" still work. A refused connect leaves the current connection alone.

2951 end-to-end tests. 32-bit build, as CH341 requires.

v2.204.0

Choose a tag to compare

@zhh198903-ctrl zhh198903-ctrl released this 27 Sep 20:47

A write field of the wrong type or range is refused, not guessed

Every field of every write endpoint was sent values of the wrong type and range. Three kinds of answer came back that should not have.

Switches read by truthiness. "x", -1 and [0] all turned these on: low_pwr, software_reset (a module reset), allow_lp_hw, bank_broadcast, squelch_method, DataPath apply / apply_immediate, and media lane switching enable / commit. They now take true or false only. An unknown action such as "rest" used to do nothing and answer ok. It is now a 400 that lists reset, low_power and high_power.

Values cut down.

  • true and 1.5 became mask bit 1.
  • PRBS pattern 300 went down as 12, and −1 as 15.
  • A raw data byte of −1 or 1FFh was written as FFh.
  • A page, address or bank of true or 1.5 became 1.

Masks are now whole numbers, patterns 0–15, data bytes 00h–FFh, and page, address and bank whole numbers. A string or null in a mask or a PRBS section used to give a 500 and now gives a 400.

Short lists padded. A short app_select list staged AppSel 0 (no Application) on every lane it left out. A short PRBS patterns list set PRBS31Q on those lanes. Both now need one entry per lane, as redirection already did. A raw write that named no page or address used to start at Lower Memory byte 0 and run through the Module Global Controls at 26–36. It now has to name both.

Every refusal happens before anything is written. The page already sends well-formed values, so this changes nothing in normal use. It matters for scripts and programs calling the API.

2941 end-to-end tests. 32-bit build, as CH341 requires.