Bad regex sort when multiple paths match

[BrknRobot]

Regex comparisons where both paths/basenames match incorrectly order the files, rather than allowing subsequent criteria to break the tie.

example where regex ordering breaks
1/a
1/a2
1/b
2/a
2/a2
2/b
pattern:
x<1>r<a>l

[SeeSpotRun]

Firstly -S x relates to basename while -S r relates to full path. So you probably want pattern r<1>x<a>l

But still there is a bug:

$ rmlint -S 'r<1>x<a>l'
# Duplicate(s):
    ls '/pwd/1/b'
    rm '/pwd/1/a'
    rm '/pwd/1/a2'
    rm '/pwd/2/a2'
    rm '/pwd/2/a'
    rm '/pwd/2/b'

Expected:

# Duplicate(s):
    ls '/pwd/1/a'
    rm '/pwd/1/a2'
    rm '/pwd/1/b'
    rm '/pwd/2/a'
    rm '/pwd/2/a2'
    rm '/pwd/2/b'

It looks like there is a problem with the regex match sorting implementation here:

rmlint/lib/preprocess.c

Lines 356 to 389 in 094fbd5

	static int rm_pp_cmp_by_regex(GRegex *regex, int idx, RmPatternBitmask *mask_a,
	const char *path_a, RmPatternBitmask *mask_b,
	const char *path_b) {
	int result = 0;
	if(RM_PATTERN_IS_CACHED(mask_a, idx)) {
	/* Get the previous match result */
	result = RM_PATTERN_GET_CACHED(mask_a, idx);
	} else {
	/* Match for the first time */
	result = g_regex_match(regex, path_a, 0, NULL);
	RM_PATTERN_SET_CACHED(mask_a, idx, result);
	}
	if(result) {
	return -1;
	}
	if(RM_PATTERN_IS_CACHED(mask_b, idx)) {
	/* Get the previous match result */
	result = RM_PATTERN_GET_CACHED(mask_b, idx);
	} else {
	/* Match for the first time */
	result = g_regex_match(regex, path_b, 0, NULL);
	RM_PATTERN_SET_CACHED(mask_b, idx, result);
	}
	if(result) {
	return +1;
	}
	/* Both match or none of the both match */
	return 0;
	}

Will look into it....