docs(tms34010-sdk): the r5.01 2ine pipeline and gspcl #13

Merged
arti merged 3 commits from opencode.energia.arti.ee/vt1200:docs/r5.01-2ine-pipeline into master 2026-11-17 04:32:37 +02:00
Collaborator

The r5.01 C tools now run end to end under the patched 2ine fork, so the
"assembler-only pipeline" caveats and the "gspac/gspcg exit 139 at teardown"
workaround are no longer true.

  • tools/tms34010-sdk/2INE-501.md — gspcl works and chains the C stages; §7
    replaces "check the output files, not the exit code" with the measured
    behaviour (0 on success, 1 for a genuine error); §1 lists the 16-bit
    stack-unmap guard among the fork's fixes.
  • tools/tms34010-sdk/OS2-501.md — the natively-run toolchain is not just
    gspa/gsplnk/gsprom any more; gspcl chains the C stages there too.
  • bad-apple/README.md, bad-apple/REPORT.md — drop the note that
    gspac/gspcg exit 139 at teardown.

The loader half is 2ine PR #1, guard commit cad0c71, merged on the fork's
main as 03c82f9:
https://git.wut.ee/opencode.energia.arti.ee/2ine-tms34010

Measured while reviewing that change, rebuilding both 2ine trees and running the
r5.01 tools this repo ships (tools/tms34010-sdk/1990-11 ...):

  • gspac -k t2.c t2.if — before: 139 after writing a valid .if; after:
    0 with a byte-identical .if (md5 fe459d38...)
  • gspcl -k -o2 -c t2.c — 0, leaves t2.obj + t2.asm; a genuine compile
    error gives 1
  • gspcl -k -o2 t2.c -z -o t2.out rts.lib flib.lib — 0, leaves t2.out

Docs only; no toolchain or harness code changes. Based on master at
6ae2358 (the two commits were cherry-picked from the branch merged in
opencode.energia.arti.ee/vt1200#1 and touch no file changed upstream since
17dfd24).

A third commit tightens the wording this PR introduced: what gspcl actually
chains (-z adds the link, libraries after it), that 3bcdfc3 is the fork's
TMS34010.md status note rather than this document's, and that the 0/1 exit
statuses are observations -- the loader keeps only the low 8 bits of an OS/2
exit code. It also records the 2026-11-16 re-verification, brings the OS/2
route's heading and lead in line with its body, and gives that route's CMake
command the CMake 4.x policy floor.

The r5.01 C tools now run end to end under the patched 2ine fork, so the "assembler-only pipeline" caveats and the "`gspac`/`gspcg` exit 139 at teardown" workaround are no longer true. * `tools/tms34010-sdk/2INE-501.md` — `gspcl` works and chains the C stages; §7 replaces "check the output files, not the exit code" with the measured behaviour (0 on success, 1 for a genuine error); §1 lists the 16-bit stack-unmap guard among the fork's fixes. * `tools/tms34010-sdk/OS2-501.md` — the natively-run toolchain is not just `gspa`/`gsplnk`/`gsprom` any more; `gspcl` chains the C stages there too. * `bad-apple/README.md`, `bad-apple/REPORT.md` — drop the note that `gspac`/`gspcg` exit 139 at teardown. The loader half is 2ine PR #1, guard commit `cad0c71`, merged on the fork's `main` as `03c82f9`: <https://git.wut.ee/opencode.energia.arti.ee/2ine-tms34010> Measured while reviewing that change, rebuilding both 2ine trees and running the r5.01 tools this repo ships (`tools/tms34010-sdk/1990-11 ...`): * `gspac -k t2.c t2.if` — before: **139** after writing a valid `.if`; after: **0** with a byte-identical `.if` (md5 `fe459d38...`) * `gspcl -k -o2 -c t2.c` — **0**, leaves `t2.obj` + `t2.asm`; a genuine compile error gives **1** * `gspcl -k -o2 t2.c -z -o t2.out rts.lib flib.lib` — **0**, leaves `t2.out` Docs only; no toolchain or harness code changes. Based on `master` at `6ae2358` (the two commits were cherry-picked from the branch merged in opencode.energia.arti.ee/vt1200#1 and touch no file changed upstream since `17dfd24`). A third commit tightens the wording this PR introduced: what `gspcl` actually chains (`-z` adds the link, libraries after it), that `3bcdfc3` is the fork's `TMS34010.md` status note rather than this document's, and that the 0/1 exit statuses are observations -- the loader keeps only the low 8 bits of an OS/2 exit code. It also records the 2026-11-16 re-verification, brings the OS/2 route's heading and lead in line with its body, and gives that route's CMake command the CMake 4.x policy floor.
opencode.energia.arti.ee added 2 commits 2026-11-17 04:30:45 +02:00
gspac and gspcg returning 139 was a real SIGSEGV, not a strange exit
code: they free their automatic data segment (DGROUP), which also holds
the running 16-bit stack, just before DosExit; lxFreeSegment unmapped it
and the 16->32 bridge faulted returning through it. Record the root
cause, the loader fix (PR #1, commit cad0c71), and that gspcl now chains
the whole pipeline. Keep the "check output files" rule for unpatched
loader builds.
The docs now describe the working r5.01 pipeline: gspcl chains the C
stages, every tool exits 0 (or 1 for a real error), and there is no
"check output files instead of exit codes" rule. The old 139/SIGSEGV
story stays in git history.
opencode.energia.arti.ee added 1 commit 2026-11-17 04:31:29 +02:00
Be precise about the driver: `gspcl` chains compile -> optimize -> codegen ->
assemble; `-z` adds the link, and everything after `-z` is a linker option, so
the libraries follow. That form was measured: `gspcl -k -o2 file.c -z -o
file.out rts.lib flib.lib` exits 0 and leaves file.out.

`3bcdfc3` is the fork's `TMS34010.md` status update, not this document's.

The exit status is an observation from the verified runs, not a loader
contract: `lxTerminate` keeps only the low 8 bits of an OS/2 exit code.

Also record the 2026-11-16 re-verification, bring the OS/2 route's heading and
lead in line with its body, and give that route's CMake command the policy
floor CMake 4.x needs.
arti merged commit 2b3e1f14f2 into master 2026-11-17 04:32:37 +02:00
arti deleted branch docs/r5.01-2ine-pipeline 2026-11-17 04:32:37 +02:00
Sign in to join this conversation.
No Reviewers
No labels
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: arti/vt1200#13