Re: A.R. Index for GDB version 6.1.91_20040719

["Eli Zaretskii" <eliz at gnu dot org>]

> Date: Mon, 19 Jul 2004 18:14:02 -0400 (EDT)
> From: mec.gnu@mindspring.com (Michael Elizabeth Chastain)
> 
> (At first I wanted to use gnat_ada.gpr_in, but then I thought:
> what if someone configures in the source directory, and what if
> configuration reaches down into testsuite/ even though testsuite/
> does not actually run on msdos).

All this does happen on MS-DOS, and therefore gnat_ada.gpr_in is not
good because the generated gnat_ada.gpr will overwrite it.

> Also my understanding is: if the ms-dos port actually uses the file, the
> name has to fit in 8.3.  But if the ms-dos port does not use the file,
> the name just has to be 8.3 unique with truncation, plus some other
> requirements like "no more than one dot".

Not really accurate: even if the file _is_ used, we only care that its
name is unique after 8.3 truncation (plus some other requirements).
This is because all DOS file-related system calls silently truncate
file names to 8.3 limits before they start to actually act on the
file.

> The test suite is not used on ms-dos so null_record.adb and
> null_record.exp are fine.  They get unpacked as null_rec.adb and
> null_rec.exp, so they wouldn't actually work, but they survive the
> process of unpacking the tarball.

And if the test suite worked for the DJGPP port, these two files would
actually work, as explained above, because when the test opens
null_record.adb, say, it will happily find null_rec.adb instead.

Note that files that are part of the configury stuff are actually
accessed on MS-DOS, since the configure script is run in its entirety,
as on Unix.