On Sun, Sep 19, 2010 at 03:11:06PM -0400, David Golden wrote: > On Sun, Sep 19, 2010 at 1:40 PM, Tim Bunce <[email protected]> wrote: > > Certainly seems like a resource problem (memory, file descriptors, etc.). > > Or perhaps the make program itself is doing something strange. > > Tim -- > > I finally looked at the errors in the report. > > My hunch is that you might be having issues with temporary files. I > don't have the tuits to decrypt your test suite, but if there is > anything that attempts to re-use directories or files between runs, > you have a huge risk on Win32 that the old file doesn't get removed > fast enough and the new file can't be created. (Perl "unlink" will > return but the OS has only scheduled deletion and hasn't necessarily > finished it.) > > I generally make sure I create a new, temporary directory with a > random name for any iterative test. And I've stopped trying to ensure > that things are cleaned up properly because it's too hard to guarantee > on Windows. (I let File::Temp do it's best to remove a directory, but > I don't worry if it fails to clean up.)
Umm, yes, good idea. (I've had similar issues with DBI tests in the past.) Though if that was the cause here I'd expect failures from other Win32 systems: 4 ok: http://matrix.cpantesters.org/?dist=Devel-NYTProf%204.05;reports=1;os=MSWin32 8 ok: http://matrix.cpantesters.org/?dist=Devel-NYTProf%204.04;reports=1;os=MSWin32 It also wouldn't explain passing reliably when not run via make. I think that, until I see some similar failures from other Win32 systems, I'm going to assume it's a local problem with Taro's setup. Tim.
