execlp/execvp needs case-correct PATH

[Thomas Wolff]

On 11.02.2015 14:28, ext Corinna Vinschen wrote:
> On Feb 10 20:14, Thomas Wolff wrote:
>> Am 10.02.2015 um 10:27 schrieb Corinna Vinschen:
>>> On Feb  9 21:49, Thomas Wolff wrote:
>>>> Am 09.02.2015 um 11:17 schrieb Corinna Vinschen:
>>>>> On Feb  9 00:04, Thomas Wolff wrote:
>>>>>> With a Windows case sensitive file system (and according mount flags
>>>>>> for /cygdrive), the PATH does not properly reflect casing of the actual
>>>>>> directories (e.g. C:\WINDOWS vs. C:\Windows, thanks MS...).
>>>>>> However, the shell finds programs anyway, like e.g. notepad.
>>>>>> The exec*p system calls, on the other hand, do not find a program in this
>>>>>> case as demonstrated by the attached test program.
>>>>>> This is in contrast to the Linux (and POSIX?) manual page which claims
>>>>>> „The execlp(), execvp(), and execvpe() functions duplicate the actions
>>>>>> of the shell in searching for an executable file …“
>>>>> ...
>>>> Sorry, I forgot one detail: I added /cygdrive/c/Windows/System32 to my path
>>>> so the shell will find it, but yet execlp does not find it.
>>> Which makes sense, given that notepad is not in C:\Windows\System32, but in C:\Windows.
>> On my systems (Windows 7 Professional/Ultimate) it’s in both C:\Windows and
>> C:\Windows\System32
>> (otherwise the shell wouldn’t have found it after adding to the path).
>>
>> However, I could resolve the issue partly by putting
>> /cygdrive/c/Windows (or ../System32) in the path *before* the bogus
>> /cygdrive/c/WINDOWS - weird, but this way exec*p works.
>>
>> With the old setting (bogus first in path), apparently/assumedly exec*p
>> somehow finds the file in /cygdrive/c/WINDOWS but then cannot start it from
>> there because of the case mis-match.
>> There’s still the inconsistency with shell behaviour.
> I found the cause.  The function searching for executables in $PATH was
> searching on the Win32 PATH variable.  The underlying conversion
> functionality treats Win32 paths with default flags.  I revamped the
> search function to iterate over the POSIX PATH variable so the
> posix=[0|1] mount flag is taken into account.  As a nice side effect,
> the search function is mow much simpler and easier to understand.
>
> I tested this new stuff in a variety of situations, but there's still
> the chance that I missed something.  So this needs a good, sturdy testing.
>
> I just uploaded new snapshots to the usual place:
> https://cygwin.com/snapshots/
>
> Please give it a try.
Excellent, thank you. I also tested with ping vs. PING vs. PING.EXE and 
behaviour is consistent.
------
Thomas

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple