Re: GNAT Modification_Time limitation

[briot.emmanuel@gmail.com]

> The ptrace approach would be much more reliable (though linux-specific), since you would know
> for instance:
> 
>     - that the compiler searched and did not find foo,ads in /first/dir
>     - found and opened /other/dir/foo.ads
> 
> so next time there is a build you can check first whether 'foo.ads' now exists in /first/dir. If that file
> now exists, you need to rebuild.
> gprbuild doesn't handle such changes on the system, it only store what it found.


Slightly out of topic (sorry): I found tup (http://gittup.org/tup/index.html) which appears to be doing
exactly what you want to achieve. It monitors file accesses but it uses a fuse filesystem for this, rather
than ptrace.

I had implemented a fuse filesystem in Ada at some point, though I do not have that code anymore. AdaCore was using that to access a database that contains all build+tests results on all possible combinations, if I remember right.