Re: [PATCH] fix crash in path.c on Windows

[Johannes Sixt <j.sixt@viscovery.net>]

René Scharfe schrieb:
> 	set a=	getenv("a")
> 	======= ===========
> 	c	c
> 	/c	c:/
> 	c/	c/
> 	/c/	c:/
> 	c:c	c:c
> 	/c:c	c:c
> 	c:/c	c:/c
> 	/c:/c	c:/c
> 	c/:/c	c\;c:\
> 	/c:c/	c:c/
> 	/c/:c	/c/:c
> 	/c/:/c	c:\;c:\
> 	/c:/c/	c:/c/
> 	/c/:/c/	c:\;c:\

Bash translates leading single-letter path components to drive prefix
notation if it invokes a non-MSYS program; and it also translates ':' to
';' if the value looks like a path list. Sometimes there is an ambiguity
and bash guesses wrong.

> @@ -387,7 +387,7 @@ int normalize_absolute_path(char *buf, const char *path)
>  	assert(path);
>  
>  	while (*comp_start) {
> -		assert(*comp_start == '/');
> +		assert(is_absolute_path(comp_start));
>  		while (*++comp_end && *comp_end != '/')
>  			; /* nothing */
>  		comp_len = comp_end - comp_start;

Junio has pointed out your thinko here. Furthermore, *all* uses of '/' in
this loop must be replaced by is_dir_sep().

But I would really appreciate if you could unify normalize_absolute_path()
with setup.c's sanitary_path_copy() because they do almost the same thing
(and the latter already works on Windows).

> @@ -438,11 +438,20 @@ int longest_ancestor_length(const char *path, const char *prefix_list)
>  		return -1;
>  
>  	for (colon = ceil = prefix_list; *colon; ceil = colon+1) {
> -		for (colon = ceil; *colon && *colon != ':'; colon++);
> +		for (colon = ceil; *colon && *colon != PATH_SEP; colon++);
>  		len = colon - ceil;
>  		if (len == 0 || len > PATH_MAX || !is_absolute_path(ceil))
>  			continue;
>  		strlcpy(buf, ceil, len+1);
> +
> +		if (has_dos_drive_prefix(buf)) {
> +			char *p;
> +			for (p = buf; *p; p++) {
> +				if (*p == '\\')
> +					*p = '/';
> +			}

IMNSHO this is a kind of normalization and, therefore, must be done by
normalize_absolute_path() (sanitary_path_copy() already does this).

> +		}
> +
>  		len = normalize_absolute_path(buf, buf);
>  		/* Strip "trailing slashes" from "/". */
>  		if (len == 1)

-- Hannes