Re: proposal: substitute-path handles foreign dir separators

[Pedro Alves <pedro@codesourcery.com>]

On Monday 20 December 2010 07:40:53, Raphael Zulliger wrote:
> First of all, your patch doesn't help for my use case. I guess (although 
> I didn't verify that) your patch only handles "relative" source file 
> names. Unfortunately, in my case, find_and_open_source is called with 
> the following argument values:
> 
>    filename: 
> "X:\\projects\\inos_projects\\inos_trunk\\os\\inos\\src\\inos.cpp"
>    dirname: 
> "/cygdrive/x/projects/inos_projects/inos_trunk/testing/hwtarget_603/"
>    *fullpath: 0
> 
> I use the following substitute-path rule:
>    set substitute-path X:\\projects\\inos_projects 
> /mnt/hgfs/projects/inos_projects/
> 
> According to my understanding, if having absolute file names in the 
> binary, the following if statement (in find_and_open_source) should 
> become true:
>    if (IS_ABSOLUTE_PATH (filename))
> which, unfortunately, is not the case because IS_ABSOLUTE_PATH 
> internally uses Unix style file name rules.
> 
> To make a long story short: I think that we still need to do the things 
> that are done by my original patch to make my use case working:
>   . Create some kind of IS_ABSOLUTE_PATH which (optionally) works for 
> all file name styles
>   . Create some kind fo IS_DIR_SEPARATOR which (optionally) works for 
> all file name styles
>   . Make sure we don't pass DOSish file names to openp (the one in 
> find_and_open_source) if running on a Unix-like system
> 
> Assuming that my statements/observations are right, how would be 
> implement these points?
> We could do similar things with IS_ABSOLUTE_PATH and IS_DIR_SEPARATOR as 
> has been done with the FILENAME_CMP macro by your patch. Having done 
> this, we probably need a third options for 
> source-filenames-matching-scheme: 'mixed'. As I mentioned in the 
> original post, it may happen that you end up with a binary which has 
> been linked from libraries that were compiled on different platforms 
> (e.g. by different developer teams). As far as I understand GDB's source 
> code handling, this would require that IS_ABSOLUTE_PATH and 
> IS_DIR_SEPARATOR need to test for both file name styles.

Yes, that sounds like a natural extension to my patch.  In
gdb/filesystem.h, we have IS_TARGET_ABSOLUTE_PATH, etc., to handle
the similar case with target paths (paths to DSOs in the target
filesystem, for example).  It sound like we need to add similar
IS_SOURCE_ABSOLUTE_PATH and friends to source.h, and make substitute-path
use them, quite like gdb/solib.c:solib_find uses IS_TARGET_ABSOLUTE_PATH,
etc.

-- 
Pedro Alves