On 05-Mar-25 14:35, Ben Burton wrote:
> I took a look at this and I'm uncomfortable with the patch that you've
> suggested.
> 
> The issue here is that I want to ensure that the installed libgcj-dev
> matches the version of gcj that is used.  This was in fact the reason
> for the latest libreadline-java upload.
> 
> The current build-depends are (gcj-3.3 && libgcj4-dev), which ensures a
> version match.
> 
> The suggested replacement of (gcj && (libgcj4-dev || libgcj-dev)) does
> not ensure a version match.  If gcj defaults to gcj-3.4, then you could
> have gcj-3.4 but libgcj4-dev (not libgcj5-dev which matches gcj-3.4).
> 
> Moreover, both libgcj4-dev and libgcj5-dev provide libgcj-dev.  This
> means that you could have gcj-3.3 (the default gcj on i386) combined
> with libgcj5-dev (provides libgcj-dev), which again gives mismatched
> versions.
> 
> If you have suggestions for working around these problems, I'd love to
> hear them.  In the meantime, I'd prefer not to apply the given patch
> because of the problems above.


Hello,

thank you for your detailed reply. 

I see your point and  I have no perfect solution for this at the moment. 
Nonetheless, I still think that it is a bad idea to make packages 
explicitly depend on a specific version of gcc. 

I am working on a port to ppc64 and there is no gcc-3.3
for this architecture, only gcc-3.4 and gcc-4.0. A large part of the
archive compiled without changes with a different default gcc.
Only very few packages depend on a specific version of gcc.

For C and C++ this is easy because the build-essential packages define
a default gcc and g++ environment.

It is certainly ugly to explicitly call 'gcj-3.3' instead of 'gcj' in 
the Makefiles. 

As a solution, I would suggest to use 'gcj' everywhere, but to introduce
a configure check which tests if 'gcj' and the installed version of 
'libgcj-dev' match, i.e. gcj==gcj-3.3/libgcj4-dev, gcj==gcj-3.4/libgcj5-dev
and gcj==gcj-4.0/libgcj6-dev.

I do not have a patch for this yet, but it should not be too difficult
to write one.

Regards
Andreas Jochens


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to