RE: [BUG] git 2.44.0-rc0 t0080.1 Breaks on NonStop x86 and ia64

[<rsbecker@nexbridge.com>]

On Saturday, February 10, 2024 9:06 PM, Junio C Hamano wrote:
><rsbecker@nexbridge.com> writes:
>
>> The diff appears to have failed because of an assumption of how paths
>> are resolved during compilation. The assumption is that files remain
>> partially qualified, which is not the case in all C compilers. This is
>> c99. My experience with gcc is that it qualifies names differently
>> than other compilers.
>
>I just found this bit of gem in t/unit-tests/test-lib.c.  I guess Visual C
falls into the
>same category as yours, while GCC and Clang are from different camps?
>
>The easiest way forward may be to build on top of it, perhaps like so, to
generalize
>the make_relative() to cover your case, too?
>
> t/unit-tests/test-lib.c | 63
+++++++++++++++++++++++++++++++++++++---------
>---
> 1 file changed, 48 insertions(+), 15 deletions(-)
>
>diff --git c/t/unit-tests/test-lib.c w/t/unit-tests/test-lib.c index
>7bf9dfdb95..69dbe1bbc7 100644
>--- c/t/unit-tests/test-lib.c
>+++ w/t/unit-tests/test-lib.c
>@@ -21,45 +21,78 @@ static struct {
> 	.result = RESULT_NONE,
> };
>
>-#ifndef _MSC_VER
>-#define make_relative(location) location -#else
>+#include "dir.h"
> /*
>  * Visual C interpolates the absolute Windows path for `__FILE__`,
>  * but we want to see relative paths, as verified by t0080.
>+ * There are other compilers that do the same, and are not for
>+ * Windows.
>  */
>-#include "dir.h"
>-
> static const char *make_relative(const char *location)  {
> 	static char prefix[] = __FILE__, buf[PATH_MAX], *p;
> 	static size_t prefix_len;
>+	static int need_bs_to_fs = -1;
>
>-	if (!prefix_len) {
>+	/* one-time preparation */
>+	if (need_bs_to_fs < 0) {
> 		size_t len = strlen(prefix);
>-		const char *needle = "\\t\\unit-tests\\test-lib.c";
>+		char needle[] = "t\\unit-tests\\test-lib.c";
> 		size_t needle_len = strlen(needle);
>
>-		if (len < needle_len || strcmp(needle, prefix + len -
needle_len))
>+		if (len < needle_len)
>+			die("unexpected prefix '%s'", prefix);
>+
>+		/*
>+		 * The path could be relative (t/unit-tests/test-lib.c)
>+		 * or full (/home/user/git/t/unit-tests/test-lib.c).
>+		 * Check the slash between "t" and "unit-tests".
>+		 */
>+		prefix_len = len - needle_len;
>+		if (prefix[prefix_len + 1] == '/') {
>+			/* Oh, we're not Windows */
>+			for (size_t i = 0; i < needle_len; i++)
>+				if (needle[i] == '\\')
>+					needle[i] = '/';
>+			need_bs_to_fs = 0;
>+		} else {
>+			need_bs_to_fs = 1;
>+		}
>+
>+		/*
>+		 * prefix_len == 0 if the compiler gives paths relative
>+		 * to the root of the working tree.  Otherwise, we want
>+		 * to see that we did find the needle[] at a directory
>+		 * boundary.
>+		 */
>+		if (fspathcmp(needle, prefix + prefix_len) ||
>+		    (prefix_len &&
>+		     prefix[prefix_len - 1] != '/' &&
>+		     prefix[prefix_len - 1] != '\\'))
> 			die("unexpected suffix of '%s'", prefix);
>
>-		/* let it end in a directory separator */
>-		prefix_len = len - needle_len + 1;
> 	}
>
>+	/*
>+	 * If our compiler gives relative paths and we do not need
>+	 * to munge directory separator, we can return location as-is.
>+	 */
>+	if (!prefix_len && !need_bs_to_fs)
>+		return location;
>+
> 	/* Does it not start with the expected prefix? */
> 	if (fspathncmp(location, prefix, prefix_len))
> 		return location;
>
> 	strlcpy(buf, location + prefix_len, sizeof(buf));
> 	/* convert backslashes to forward slashes */
>-	for (p = buf; *p; p++)
>-		if (*p == '\\')
>-			*p = '/';
>-
>+	if (need_bs_to_fs) {
>+		for (p = buf; *p; p++)
>+			if (*p == '\\')
>+				*p = '/';
>+	}
> 	return buf;
> }
>-#endif
>
> static void msg_with_prefix(const char *prefix, const char *format,
va_list ap)  {

This looks like a good plan.

--Randall