CBM_EXTRACT_BUDGET is documented as a byte budget but is passed as a parse timeout

[BobbieBarker]

src/pipeline/pipeline_internal.h:26

/* Maximum byte budget for tree-sitter extraction per file */
#define CBM_EXTRACT_BUDGET 5000000

It is not a byte budget. It is passed as the timeout_micros argument at eight
call sites (pass_calls.c:779, pass_definitions.c:675 and :834,
pass_k8s.c:655 and :673, pass_parallel.c:1167, pass_semantic.c:667,
pass_usages.c:356), and internal/cbm/cbm.h:734 documents that parameter as:

// timeout_micros: per-file parse timeout in microseconds (0 = no timeout).

So the value is 5 seconds of CPU time. internal/cbm/cbm.c:2167 confirms it:

budget.cpu_deadline_ns = cbm_thread_cpu_time_ns() + budget_ns;
budget.wall_ceiling_ns = t0 + budget_ns * CBM_PARSE_WALL_CEILING_FACTOR;

and internal/cbm/cbm.c:299 already names it correctly:

#define CBM_PARSE_WALL_CEILING_FACTOR 12ULL /* ~60 s ceiling for the 5 s CPU budget */

So the header comment is just wrong. Renaming the constant is a wider change, but the comment can be corrected on its own.

Line numbers are against 5958f546.

[DeusData]

Thank you so much @BobbieBarker for this careful report. Tracing all eight call sites through to cbm.h and the CPU-deadline code in cbm.c made it quick to confirm.

You're right: CBM_EXTRACT_BUDGET is a per-file tree-sitter parse budget in microseconds of the parsing thread's CPU time (5 s), with a wall-clock ceiling of about 60 s as a backstop. It was never a byte limit. Every caller already uses the value correctly as timeout_micros, so only the documentation was wrong.

#2358 corrects the comment on the constant, and also the timeout_micros description in cbm.h, which still said "parse timeout" and had drifted away from cbm_extract_file(). As you suggested, we've kept the rename out of this change so the fix stays small. Thanks again for the precise pointers!

[BobbieBarker]

In the future, if you want me to implement && resolve any issues I open, I'll be happy to do so. Just let me know that you agree with the framing of the problem, and I can handle the rest.