To check whether or not two dirs are the same without stat, one could cdhir to each
dir (from the base dir) and then getcwd. As far as I know this always returns the
canonicalized path, therefore makes the two path comparable. This can work on
But let's backtrace a bit:
As far as I can tell, the sequence of events was as follows:
1. seyko commits "mormalize_inc_dirs"
2. I said (and now adding: on windows 8.1 with msys2 and gcc 5.2.0):
Prior to this commit, when building tcc itself (configure and make), after successfully
building the tcc executable (on windows for windows), it starts building libtcc1.a using
the just-compiled tcc.exe, and succeeded.
This process now fails for me (missing stddef.h for libtcc1.a) ...
BTW this commit breaks tcc compiling itself afterwards.
WinXP SP3 32bit, MinGW(tcc self compilation test)
Using the compiled tcc to recompile tcc itself fails with windows.h not found.
I have to reverting this commit to recover.
So far, from these two comment, it's apparent that the newly built windows tcc
executable does not resolve dirs the same as it used to. It's unclear whether
other platforms are affected.
4. seyko responds to Toy Tam:
Anyone know solution for stat? From Google:
"stat" not working on Windows XP using v14_xp platform ...
stat() does not work for me on Cygwin/Windows - C / C++
Portable way to check if directory exists?
5. (now) A discussion goes on how to compare dirs on XP.
So, do we know for a fact that this is the actual problem? I.e. that the current
dir resolution failure is due to missing code which needs to check if two dirs
If such required dir comparison was missing from the code, then I'd expect that
other platforms are affected as well. But as far as I can tell there wasn't indication
Though it's possible that other platforms are affected, but the issue doesn't affect
the tcc build process on those platforms, but it does on windows.
However, how come the old code did not need to compare such two dirs, and still
worked correctly? If the old code was working correctly without needing to compare
two actual dirs, despite it not being "n[m]ormalized", maybe it's still the better
solution because it depends less on features which are harder to use reliably on
On Saturday, November 7, 2015 3:59 PM, Edmund Grimley Evans <address@hidden> wrote:
> Hum! You can:
> fopen("dir-to-test/__deleteme__", "w");
> If it succeeds, dir-to-test exists then call fclose and
> remove("dir-to-test/__deleteme__").
As I understand it, the problem to be solved is not determining
whether a directory exists, but determining whether two directories
are the same (even though the paths might look quite different). You
can do that under POSIX by comparing st_dev and st_ino.
I suppose you could check that neither PATH1/RANDOM_NAME nor
PATH2/RANDOM_NAME exists, then create one of them and see if the other
now exists. But in most cases the directories are not writable!
Does GCC have an alternative implementation for systems without stat?
Edmund