Re: [Cgdb-devel] Cgdb port for Cygwin

[Ronald Landheer-C. <bly...@us...>]

On Mon, Nov 24, 2003 at 01:47:00PM -0500, Bob Rossi wrote:
> > I've made the first set of changes to my local sandbox to allow out-of-tree 
> > building and am about to commit them (or rather: will commit them tomorow 
> > morning) to the cygwin package's branch. 
> Great. I am glad to hear it!
> 
> > Are there any objections against me
> > removing the generated files (Makefile.in, configure, etc.) from the branch?
> > They tend to bloat the CVS contents and, for the moment, do not exactly 
> > reflect my changes because they're re-created with a different set of 
> > Autotools than I will use to make the final package.. (IMHO, no generated
> > files should exist in CVS anyway..).
> The original idea was that the user would not need to have autoconf/automake
> installed in order to get CGDB to build. They could just do ./configure
> && make. The good thing about that is the user does not have to go get
> the versions of the tools we are using in order to get CGDB to build.
> They only need that software in order to change the build system.
> 
> By removing those files, we would be complicating what the user has to
> do in order to use CGDB.
> 
> What does everyone think the best way to go is?
Personally (but I'm repeating myself), I think a user that wants to build from 
source should take the source tarball - either the one provided for his/her 
distribution (which should contain the generated files, of course) or the
one released by the project if one for his/her distribution doesn't exist.
The released source tarballs should, of course, contain the generated files.

A developer working on the project should be able to generate the files, and
may or may not do so with the same version of the tools as the person that
makes the distribution. I, for one, did most of the porting I did so far on
the Cygwin branch on a Gentoo box with a slightly outdated set of Autotools,
which doesn't change much functionally (as the configury files for cgdb are
more or less garden variety) but it does make a rather significant difference
in the generated files, which is transient as the files will finally be 
generated for Cygwin, on Cygwin, with the latest Autotools. The result is a 
rather bloated patch (as the one for the previous release) with only very few
man-made changes. For info: the diff with all files (generated and man-made)
is 253K (7908 lines) in size, w/o ChangeLog entries; the one without the
generated files (aclocal.m4, configure and Makefile.in) is 3.7 K (111 lines).
The difference of almost 7800 lines in a patch - all of which are generated - 
seems rather significant to me, especially if someone wants to review the 
changes I made on the branch to merge them into HEAD: 7800 lines of clutter
could easily hide a few lines of bugs.

The reason I don't like having generated files in CVS is not only that it
bloats the repository - which from an administrators perspective is reason
enough - but that it also bloats man-made changes and tends to obfuscate 
those changes. I don't know how you review changes, but I generally look
at the patch before applying it and 7800 lines of generated clutter certainly
makes that more difficult.

That said, I re-iterate that removing the files from the branch has no impact
on HEAD, and as the branch is intended for a Cygwin release which, itself,
will contain the necessary (generated) files, it will have no impact on 
Cygwin-cgdb users either.

rlc


-- 
Most people don't need a great deal of love nearly so much as they need
a steady supply.