cscope

Hans-Bernhard Broeker wrote:

>On Sun, 3 Oct 2004, Neil Horman wrote:
>
>  
>
>>It effectively ensure that if a file is a link, that it points 
>>to a readable file before it adds it to the search list.  
>>    
>>
>
>I don't see how this could have any effect, given that the struct stat
>you're checking was obtained by lstat(), not stat().  I don't think
>S_ISLNK() can ever be true in that case, can it?
>
>Also note that this patch may be rendered unnecessary by the changes to
>issrcfile() already in cscope CVS.
>
>  
>
 From my reading of lstat, it provides filesystem data on the link 
inode, rather than on the inode of the file the link points to, as with 
lstat.  And yes, S_ISLNK can and is true if lstat is used on a symlink 
(I tested that).

Please don't misunderstand me, I don't want to claim this is a fix to 
the problem that was reported to me (I'm still in the process of 
root-causing that), I just thought this might be a worthwhile check to 
add since (I think) the code as it stands now, doesn't add regular files 
that are unreadable, but tries to add readable symlinks that point to 
unreadable files.

Neil

-- 
/***************************************************
 *Neil Horman
 *Software Engineer
 *Red Hat, Inc.
 *nhorman@...
 *gpg keyid: 1024D / 0x92A74FA1
 *http://pgp.mit.edu
 ***************************************************/

Hans-Bernhard Broeker wrote:
> On Sun, 3 Oct 2004, Neil Horman wrote:
> 
> 
>>It effectively ensure that if a file is a link, that it points 
>>to a readable file before it adds it to the search list.  
> 
> 
> I don't see how this could have any effect, given that the struct stat
> you're checking was obtained by lstat(), not stat().  I don't think
> S_ISLNK() can ever be true in that case, can it?
> 
> Also note that this patch may be rendered unnecessary by the changes to
> issrcfile() already in cscope CVS.
> 

Hans-
	I checked cvs for cscope changes that may have repaired this problem, 
but unfortunately the problem persists there.  I don't have it 100% 
sorted out, but I'm fairly confident that this bug is stemming from a 
discrepancy between the source file list (srcfiles) and the database 
file.  In the crossref function, it would appear that every file which 
is in the source file list, gets an entry in the cscope database which 
at a minimum consists of its filename, two newlines and a tab.  This 
minimal entry is not added to the database, however, in the event that 
the file is unreadable (not a regular file, a link that points to a 
non-existant file, etc).  This descrepancy seems to skew the database 
file index, which confuses cscope into thinking symbols are 
used/defined/etc in files which they are not.  I think the solution to 
this should be either prevent non-readable files from being added to the 
source file list (a generalization on my previously proposed patch),  or 
a change which forces crossref() to add the aformentioned minimal entry 
to the database for every file in srcfiles.  The attached patch 
implements the latter idea.

Thoughts and comments appreciate.  Thanks!
Neil


	

-- 
/***************************************************
  *Neil Horman
  *Software Engineer
  *Red Hat, Inc.
  *nhorman@...
  *gpg keyid: 1024D / 0x92A74FA1
  *http://pgp.mit.edu
  ***************************************************/