:P test failure

:P test failure

[Peter Stephenson]

This doesn't work on my machine:

 # This is a bit brittle as it depends on PATH_MAX.
 # We could use sysconf..
 bad_pwd="/${(l:16000:: :):-}"
 print ${bad_pwd:P} | wc -c
0:modifier ':P' with path too long
>16002

The modification fails with error and output

zsh: path expansion failed, using root directory
2

pws

Re: :P test failure

[Daniel Shahaf]

Peter Stephenson wrote on Thu, 26 Mar 2020 21:27 +0000:
> This doesn't work on my machine:
> 
>  # This is a bit brittle as it depends on PATH_MAX.
>  # We could use sysconf..
>  bad_pwd="/${(l:16000:: :):-}"
>  print ${bad_pwd:P} | wc -c
> 0:modifier ':P' with path too long
> >16002  
> 
> The modification fails with error and output
> 
> zsh: path expansion failed, using root directory
> 2

The behaviour you see is exactly what the test's _expected_ output was
before 45583/0004.  Is there any chance that you ran HEAD's tests on an
old zsh binary by accident?

What's the value of errno?  (Patch at the end for your convenience)

Cheers,

Daniel

diff --git a/Src/utils.c b/Src/utils.c
index 69885fed3..8a7558700 100644
--- a/Src/utils.c
+++ b/Src/utils.c
@@ -1030,8 +1030,9 @@ xsymlink(char *s, int heap)
     if (*s != '/')
 	return NULL;
     *xbuf = '\0';
+    errno = 0;
     if (!chrealpath(&s, 'P', heap)) {
-	zwarn("path expansion failed, using root directory");
+	zwarn("path expansion failed (%e) using root directory", errno);
 	return heap ? dupstring("/") : ztrdup("/");
     }
     return s;

Re: :P test failure

[Peter Stephenson]

On Thu, 2020-03-26 at 22:21 +0000, Daniel Shahaf wrote:
> Peter Stephenson wrote on Thu, 26 Mar 2020 21:27 +0000:
> > This doesn't work on my machine:
> > 
> >  # This is a bit brittle as it depends on PATH_MAX.
> >  # We could use sysconf..
> >  bad_pwd="/${(l:16000:: :):-}"
> >  print ${bad_pwd:P} | wc -c
> > 0:modifier ':P' with path too long
> > > 16002  
> > 
> > The modification fails with error and output
> > 
> > zsh: path expansion failed, using root directory
> > 2
> 
> The behaviour you see is exactly what the test's _expected_ output was
> before 45583/0004.  Is there any chance that you ran HEAD's tests on an
> old zsh binary by accident?

Yes, rebuilding does indeed fix it.  Somehow I managed to get it built
with my patch in but not yours --- probably the result of a merge.

Cheers
pws