Re: Right place for possible bug in mingw32-make

[Jeroen Cranendonk]

Hi!
See inline replies :)

On 4/19/2010 9:41 AM, Eli Zaretskii wrote:
> > Date: Fri, 16 Apr 2010 02:07:02 +0200
> > From: Jeroen Cranendonk<address@hidden>
> >
> > empty test.txt in folder.
> > Makefile in same folder with:
> > all:
> >       "cat.exe" test.txt test.txt test.txt test.txt test.txt test.txt>>
> > (27440 bytes in this example, for just the arguments).
> > cat.exe in the same folder.
> >
> > Btw, the quotes around cat.exe seem to matter.
> >      
> Yes, the quotes cause the shell to be invoked to handle the command,
> instead of invoking cat.exe directly.
>    
Ahh, interesting. Is there a way to not use quotes and handle spaces in 
folders gracefully?
Probably a noob question, sorry :) I'm just using makefiles from 
existing projects, not an expert on them ;)

> > If I now call 'mingw32-make' in the folder, the output ends with:
> >
> >    test.txt test.txt test.txt test.txt test.txt test.txt test.txt
> > test.txt test.tx
> > t test.txt
> > cat.exe: est.txt: No such file or directory
> > cat.exe: tet.txt: No such file or directory
> > cat.exe: testtxt: No such file or directory
> > mingw32-make: *** [all] Error -1
> >
> > Notice the mangled names there, dissapearing characters.
> >      
> Hm, I cannot reproduce this.  Does the number of characters have to be
> _exactly_ 27440, or will larger numbers do as well?  I tried to hit
> the exact number, but I'm not sure I succeeded.
>    
The number of characters does not have to be 27440 exactly :) That was 
just what I happend to end up with after a lot of copy pasting.
There seems to be a disspearing character at exactly every 8k'th 
position in the arguments (not including the .exe).

Mind, the test I gave you was a quickly thrown together one, my actual 
problem occurs with java.exe, which logs in more detail what went wrong, 
so in that case I could really find the positions, with cat you can only 
see some files are misnamed.

> Anyway, on my system, I get 2 kinds of results:
>
>    . with a port of Unixy shell sh.exe somewhere on PATH, I get the
>      full unmangled output.
>
>    . if I remove sh.exe from PATH, and Make uses cmd.exe as the shell,
>      I get this self-explanatory error message:
>
>        The input line is too long.
>        make: *** [all] Error 255
>
> With cmd.exe, the error message is expected, because cmd.exe is
> documented to have a limit of 4K characters in input lines.
>
> My make.exe is built from patched sources, but I tried with the MinGW
> build from make-3.81-20090914-mingw32-bin.tar.gz, and I get the same
> results as shown above.
>
> What OS are you running this on?  Do you have some sh.exe on PATH, and
> if you do, what port of which Unix shell is that?
>
>    
Hmm, OS may be a big factor here, running Windows 7 64 bit here!
Should've mentioned that sooner, sorry :)

There is no sh.exe anywhere, it's a pretty minimal mingw install.
Mayby with windows 7 cmd.exe they decided to allow longer arguments?
And mayby that's not supported well in make-w32 since noone could test 
it before?
OR mayby windows7 cmd.exe is broken, now that'd be fun :)

If XP will give a too long error anyways, I need to get back to the guy 
who writes the makefiles though, since this won't work there anyways.

Let me know if I can be of more assistance :)

Cheers!
Jeroen