The suggestion itself was missing in Jack's message.

It was that the pkg containing the
symlinks in %p/bin could, for future gcc4X pkgs,
be just called "gcc".

Existing fink pkgs would obviously not be affected,
since all their deps and bdeps involve gcc4X.

>> The suggestion does have the slight disadvantage
>> that new pkgs, if they want to depend on a
>> specific version of gcc, have to Depend on gcc4X-compiler,
>> and put %p/lib/gcc4.X/bin  first in PATH

But this is also an advantage, since it avoids Bdeps on
(mutually conflicting) gcc4X pkgs _ thus allowing even
parallel builds of on one side a pkg depending on gcc45 and
on the other side on gcc46.
>>
>> But existing pkgs don't have to be modified
>> (provided the existing gcc4X pkgs are not refactored !),
>> assuming just that in your upcoming gcc-4.4.4 pkg
>> you add "gcc" itself to the Conflicts and Replaces.
>>
>> This would still cause slight trouble with the pkgs still
>> depending on gcc42 or gcc43 , users having to manually
>> install that dep.. I'd think there are few such pkgs,
>> and that there are anyway probably reasons to upgrade
>> most of them.
>> But some seem to have reasons (dx, pdftk).
>> Probably it would be worth, for those 2 pkgs, to add also
>> to gcc42 and gcc43 a Conflicts/Replaces for "gcc"...
>> Hopefully very few users still need those old versions,
>> so few would have to re-compile ...
>> And the same thing would anyway have to be done
>> oterwise for the upcoming gcc-4.6 for those pkgs..
>>
>> This is "biting the bullet" once; after this, no more trouble
>> ever of this sort.
>>

On 5/3/10 12:12 PM, Jack Howarth wrote:
>> ps My main concern is that developers will have buyer's
>> remorse if they automate the BuildDepends on gcc such
>> that the newest FSF gcc is always used and they suddenly
>> discover package X is incompatible with the compiler
>> changes or tickles a bug in the compiler.
Of course.
As a rule, pkgs should depend, as said above, on a fixed version,
and set PATH appropriately.
(but _ for pkgs I maintain, there are a couple (say saclib,
qepcad) where I might be tempted to examine the possibility
to just let them depend on "gcc"

>> Also, we will
>> have to propose some convention for patch replacement
>> such that a package that BuildDepends on gcc automatically
>> checks its version and sets LDFLAGS=-L%p/lib/gcc4.x/lib
>> based on the version returned. In short, this is not
>> a trivial undertakening and everyone will have to buy
>> in.
Don't understand this.
Existing pkgs should not be touched.

>
On 03 May 2010, at 19:07, Alexander Hansen wrote:
> Perhaps I'm missing something here (maybe it was in one of the other
> threads) but how is it possible _not_ to have to update the Depends:  
> of
> packages that use gccX when X is changed?
gccX continues to exist, even if a "gcc" pkg comes up.
>
> BuildDepends are well and good, but what about the libraries from
> FSF-gcc that get linked? (i.e. why one reason we have gccX-shlibs for
> multiple X)  Are they going to be refactored to maintain their
> install_name across versions (and then what about ABI compatibility)?
> Or will there no longer be any linkage?
The -shlibs splitoffs are not affected by this suggestion
>
> Otherwise, it seems like we're still going to have to have packages
> Depend on a particular gccX-shlibs.
sure !

Jean_francois


------------------------------------------------------------------------------
_______________________________________________
Fink-devel mailing list
[email protected]
http://news.gmane.org/gmane.os.apple.fink.devel
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-devel

Reply via email to