Beagle SCM

PTR-009: abc/URI static Buf4096 needs an API change (carved from PTR-001)

PTR-001 split out the two abc/URI.c static u8 Buf[4096] because deleting them is a real DESIGN FORK, not a mechanical refactor: URIMergePath/URIRelativePath return slices that VIEW into the static buffer and the caller (URIAbsolute/URIRelative, public alias n) does NOT copy them out — abs->path keeps pointing into the buffer after return. a_carve can't replace them: BASS rewinds at the call() boundary, so a carve inside the callee dangles the returned slice. GATED — abc is core, surface the diff.

Input

Context

• abc/URI.c:301 static u8 URIMergePathBuf[4096] — URIMergePath (:305) memcpy's base-dir+rel into it, runs RFC-3986 remove_dot_segments with in/in_end/out cursors (in+=3, *out++, out[-1]), and returns {Buf, out} into result (= abs->path). Sole caller :444.
• abc/URI.c:140 static u8 URIRelPathBuf[4096] — URIRelativePath (:162) builds ../-climb + tail segments into it, returns {Buf, out} into rel->path. Sole caller :253.
• Both results outlive the call: URIAbsolute/URIRelative return with abs->path/rel->path STILL viewing the static buffer (no u8csDup out). Non-reentrant: a second call clobbers a live first result. External caller sniff/SUBS.c (n(&abs,&dbase,&rel)) relies on the current contract.

Goals

• Delete both static Buf[4096]; the computed path must live in caller-owned storage with a clear lifetime.
• Prefer an abc/PATH segment-walk (PATHu8sDrainNE + a_carve/gauge) over the in/in_end/out cursor for remove_dot_segments — no manual RFC parse in C (ABC).
• Keep URIMergePath/URIRelativePath/URIAbsolute/URIRelative behavior byte-identical (abc/test/URI table + RESOLVEURItest stay green).

Constraints

• GATED: abc is core + public (n) — surface the diff, never autoland; the chosen API shape needs human review.
• This is an API decision, not a free refactor: the scratch/result storage must be threaded from a caller that outlives the use, which changes the public URIAbsolute/URIRelative signature OR introduces a per-uri owned path buffer. Pick the shape WITH the maintainer before coding.

WIP

Design decisions

• Candidate A: add a caller-provided u8s scratch (or Bu8) param to URIAbsolute/URIRelative, carve it at the top call site (SUBS.c, tests); callee feeds the computed path in, result views the caller's buffer. Ripples to sniff/SUBS.c + abc/test/URI.
• Candidate B: give uri its own owned path-scratch slot (struct grows) so the result can view abs's own storage; revisits "uri struct = 8 slices + separate text" (URI-004) — likely rejected, but note it.
• Either way remove_dot_segments becomes a PATH-segment walk into a gauge, not the out++/out[-1] cursor.

TODOs

• Agree the API shape (A vs B) with the maintainer BEFORE coding.
• Repro-first: lock abc/test/URI + RESOLVEURItest outputs.
• URIMergePath → PATH-segment remove_dot_segments into caller storage; drop URIMergePathBuf.
• URIRelativePath → caller storage; drop URIRelPathBuf.
• Thread the storage through URIAbsolute/URIRelative + update sniff/SUBS.c + tests.
• Build + ctest + URI fuzz; surface diff for review.

Blockers and bummers

• Public-signature change collides with any open URI-00x work on abc/URI.c; coordinate before forking.

Outcome

• Ruled Candidate A, u8s out variant. Landed b0455595 (abc sub 5f84502a):

  URIAbsolute/URIRelative take a caller-owned u8s out (callers pass u8bIdle(buf) over a_pad); the result VIEWS the written prefix via a gauge (u8gOf/u8gFeed/u8csMv(result, u8gLeftC(g))) so out must outlive it.

• Both static Buf[4096] deleted; remove_dot_segments → PATHu8sDrain

  segment-walk; merged concat stays internal BASS a_carve. 397/397, ASAN-clean.