Erlang/OTP source reference

Clause tracks the official maint-29 branch. The reviewed pin lives in references/otp-pin.cmake:

Revision e14591f196d4eacc1946e7073e0ec27f72df2a12 (OTP 29.1.2, upstream commit dated 2026-10-26), refreshed on 2026-10-27.
Installed oracle used for goldens and audits: OTP 29.1.1 / ERTS 17.1.

references/otp is an ignored checkout. Normal builds and tests use project-owned goldens and need neither the checkout nor installed OTP. Configuration and tests never fetch or advance the reference. Live audits are opt-in with -DCLAUSE_OTP_AUDITS=ON; see fixture regeneration.

OTP source, copied helpers and OTP-generated headers never enter Git. Committed test inputs are locally authored; committed goldens are observations of those inputs. See validation.

Refresh procedure

Run at the start of OTP-dependent work.

Check git -C references/otp status --short and preserve local work. Fetch: git -C references/otp fetch https://github.com/erlang/otp.git refs/heads/maint-29:refs/remotes/origin/maint-29. Review git -C references/otp log -1 origin/maint-29 and changes since the pin, especially grammar, scanner, preprocessor and corpus inputs.
Fast-forward the clean local branch: git merge --ff-only origin/maint-29 (first setup: git switch --create maint-29 --track origin/maint-29). Use LF checkout bytes (core.autocrlf=false) so checksums match on every host; on an existing Windows checkout normalize only manifest-listed text files and erl_parse.yrl, and keep upstream binary fixtures unchanged.
Generate the compiler header from the same revision (from the project root): perl references/otp/erts/emulator/utils/beam_makeops -compiler -outdir references/otp/lib/compiler/src references/otp/lib/compiler/src/genop.tab. On Windows normalize only the generated beam_opcodes.hrl to LF.
Update the pin and verify every path in tests/fixtures/parser/phase6/otp.tsv. Refresh hashes only after reviewing real upstream changes. If erl_parse.yrl changes, review tests/fixtures/parser/grammar.tsv, its fixtures and phase6/coverage.tsv; every ordinary production needs a measured witness.
Configure with CLAUSE_OTP_AUDITS=ON and CLAUSE_OTP_SOURCE_ROOT set to the checkout. Run the audit tests and tests/compiler/patternmatch/regenerate.py --corpus all --check. Record the revision and outcomes below; report drift instead of regenerating to hide it.

Pin history

Historical records keep the revision they were made against.

Date	Event	Revision
2026-09-19	Parser validation against OTP 29.1 tag	`751f87b703fe5948607d08e82599ce644b772e76`
2026-09-28	Switched to `maint-29`; grammar and ten corpus files identical	`21776803ecd11f5fa948732c0ec66b8f325dedfc`
2026-10-01 – 10-03	Re-fetched before each pattern/guard step; unchanged	same
2026-10-03	Plan 11 step 1: 14 audit tests pass, 19 corpora reproduce	same
2026-10-14	Refreshed to OTP 29.1.2: grammar, scanner, preprocessor and `phase6/otp.tsv` hashes unchanged (`genop.tab` comment only); 17 audit tests pass and 19 corpora reproduce under oracle 29.1.1; the revision recorded by executable, lint, program and printing goldens updated (outputs unchanged, `external_terms` keeps its atom-order-dependent map output)	`d199a49a01a2079afb6f2c5c9ae047cd4555e68d`
2026-10-24	Refreshed; upstream changed only `.github/` workflows (32 files), so grammar, scanner, preprocessor, corpus and `phase6/otp.tsv` inputs are byte-identical and audits were not rerun; the revision recorded by goldens updated	`9710673e5916164dff9616327df7de39f1aa75a9`
2026-10-27	Refreshed; upstream changed only `.github/` workflows (31 files), inputs byte-identical, audits not rerun; the revision recorded by goldens updated	`e14591f196d4eacc1946e7073e0ec27f72df2a12`