[Bug]: build-ss3-warnings only counts gcc source lines numbered 1000-9999

[Santoshkumarpuppala]

Describe the bug

build-ss3-warnings.yml:79 (at 82f92f7) counts warnings with grep(pattern = "^ [0-9]", x = txt). gcc right-aligns line numbers in a 5-character field, so only source lines numbered 1000-9999 match: grepl("^ [0-9]", c(" 121 |", " 1947 |", "17560 |")) is FALSE TRUE FALSE. In the generated ss3.cpp (40,109 lines), 1000-9999 is DATA_SECTION code and FUNCTIONs start at line 16627. A new unused-variable warning in a FUNCTION passes the job with "Acceptable number of warnings: 0".

A counted warning fails the step on n_warn[1,1] applied to an integer (line 88, incorrect number of dimensions), not on the stop() at line 91. Line 95 reduces to n_warn < 0, so "decreased" never prints.

To Reproduce

I ran ADMB 13.2 tpl2cpp on 82f92f7, added one unused int per case to the tpl sources, compiled with the CI command under g++ 16.2, put the output into the warnings.txt layout from run 36063469992, and ran the R step unchanged:

unused variable added in	ss3.cpp line	gcc warnings	job
nothing (baseline)	-	0	pass, count 0
SS_readdata_330.tpl LOCAL_CALCS	1947	1	fails on the R error
get_natmort()	17560	1	pass, count 0
write_msg() in GLOBALS	121	1	pass, count 0

The -Wmisleading-indentation warning quoted in #561 (ss.cpp:34714) also passes. Counting ": warning:" lines instead counts each of these. Debian gcc 12.2 (C front end) prints the same margins; the CI image johnoel/admb-13.2:linux was not run. The latest run (36063469992) had no gcc warnings.

Expected behavior

Any new warning in SS3 code fails the job with "Increased number of warnings".

Which OS are you seeing the problem on?

Linux

Which version of SS3 are you seeing the problem on?

main at 82f92f7

Additional Context

The regex came in with #561 (312b64c). Happy to open a PR that:

1. counts ": warning:" lines;
2. starts at warn_line+2: line 76 skips the first gcc line, which for a file-scope warning is the warning itself (an unused static int at ss3.cpp:20 counted 0 with only the regex fixed);
3. drops lines 87-89;
4. compares against the reference file's warning count instead of length(ref), which is 3 (with only 1 and 3, one new warning passed).

With all four, the baseline passed and every warning above reached the stop().

[iantaylor-NOAA]

@Santoshkumarpuppala, thanks for reporting this issue.
I'd be happy to review a pull request which addresses the issue.
Can you tell us more about how you came across the problem in the first place?

[Santoshkumarpuppala]

I came across it while reviewing CI checks in public scientific and government repositories for gates that can pass without checking what they claim, not from running SS3 or working on the model.

I'll open the PR with the four changes listed in the issue, plus the dead "decreased" branch at line 95. The regex alone does not close it: with ": warning:" counted but the threshold still length(ref), one new warning prints "Acceptable number of warnings: 1".