Skip to content

AlxLib 1.3.2

Choose a tag to compare

@AlexisVon AlexisVon released this 24 Oct 01:04
· 5 commits to main since this release

Patch release — the child process's whole tree, the debug hook's scalar position and the send-queue depth, and the first Windows binaries.

process_ctrl owns the child's tree (1.3.2): every start now runs the child in a tree of its own — POSIX a process group (setpgid on both sides of the fork), Windows a job object created for every start rather than only when process_limits asked for one — and kill() / destruction signal that group / terminate that job, so what the child started goes with it, including the descendants left behind by a child that had already exited. terminate() still signals the direct child only; a descendant that calls setsid() or breaks away from the job is beyond the platform's reach, as documented. status() stops reaping: it peeks with waitid(…, WNOHANG | WNOWAIT) and leaves an exited POSIX child a zombie until kill(), the only reaping point now — that is what keeps the pid, and with it the process group id kill() signals, reserved to the object, so a late kill(-pgid) can never fire into a recycled group. Hosts that relied on status() reaping call kill() (the destructor already does); the outcome keeps coming back through status(), and the child table shows a defunct entry until then.

hook_info gains scalar positions and a path reference: wkdt_row() / wkdt_col() read the state's position slot without building the varmap that wkdt_pos() allocates (≈ 250 ns a statement — the GUI host measured a 2.5× debug mode around it), and wken_path(_e) returns the resolved full path by reference, empty for the root, where wken_file() returns a copy and the position's file field only the basename. Non-virtual, no vtable slot, no hook_info field, no opcode or vtype change. comm_ex gains the queue depth: comm_ex(transmit*, uint_64 _queue_limit = 0x80) forwards comm's existing send-queue bound to the constructor, so a host pushing large snapshots can widen it; the default and every send/receive semantic are unchanged.

The Windows side builds and ships: the MSVC Release|x64 solution builds again and all four suites go green — the toolhelp32 enumeration names PROCESSENTRY32W / Process32FirstW / Process32NextW, which are the only spellings the Windows SDK ships, and the <charconv> include gate follows the library's own CPP_VERSION macro instead of __cplusplus, which MSVC leaves at 199711L without /Zc:__cplusplus. This is the first release to carry a win-x64 package, and it is DLL-based: the four DLLs with their import libraries (the lib/*.lib members name only the DLLs they bind to), the public headers and the docs. No static library is shipped on Windows — the DLL shape is the one supported form, keeping /MD and /MT out of the picture — and the PDBs travel as a separate archive.

Migration notes: no opcode or AST change anywhere in this, so vtype is untouched and .axp products need no recompilation; no symbol changed its home, the SONAME stays .so.1.3, and the library version moves 1.3.1 → 1.3.2 — a patch. One host-visible behaviour change: process_ctrl::status() no longer reaps, so a host that waited on it for the child's disappearance should call kill(). The migration record is in doc/scpt/design.md §10.1.

Build packages — AlxLib-1.3.2-linux-x64-g169f9ef-slim.tar.gz (+ .sha256): the public headers, the merged static library (lib/libalxlib.a, debug info stripped) and the notices; the shared libraries and per-module archives stay in the developer package and are not published. AlxLib-1.3.2-win-x64-g169f9ef.zip (+ .sha256): the four DLLs, their import libraries, the headers and the docs, with AlxLib-1.3.2-win-x64-g169f9ef.pdb.zip (+ .sha256) carrying the PDBs.