Re: why cgen/cpu and not cgen in gdb_5_2_1-2002-07-23-release

[Doug Evans <dje@transmeta.com>]

Andrew Cagney writes:
 > > I just checked out gdb_5_2_1-2002-07-23-release from the cvs tree.
 > > 
 > > Question: Why are the cgen cpu files there but not cgen?
 > 
 > Same reason GDB doesn't include autoconf, automake, gettext, bison, and 
 > many other tools used to create generated files.  Not needed.

I recognize this.
But cgen isn't autoconf.  gdb/configure.in isn't shipped with autoconf.

I'm wondering if more changes are required or different rules are at play.
That's all.

Methinks apps shipping the .cpu files in src/cgen/cpu without cgen is fragile.
How fragile I dunno, but it is suspect.  Ergo my question.
[N.B. I'm not suggesting not shipping .cpu files.
Nor am I suggesting shipping the cgen *.scm files.
I'm just questioning the current situation.
As an example, one could move the .cpu files to a different dir.]

If I upgrade to autoconf 2.15, or some such, I don't expect any fundamental
change to gdb.  If I grab a copy of cgen off the net, it'll come with
the .cpu files.  All of a sudden my gdb 5.2 is now supporting the
foo and bar insns of the baz cpu (assuming one configures the tree with
--enable-cgen-maint or some such).
I suppose we could have two different cgen releases,
one with .cpu files (*1), one without.  [Or, for completeness' sake, cgen
could be instructed to use the .cpu files that came with the app, rather
than the ones that came with it, but that's clearly rather fragile.]

(*1): There's also .opc files.  I'm using ".cpu files" as a catch-all.
[One can certainly argue .opc files should live in opcodes, but that's
another discussion.]

Also, maybe now's the time to add version numbers to .cpu files.
That is also another discussion.