gnustep-dev
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Cutting a gnustep-gui release


From: Eric Heintzmann
Subject: Re: Cutting a gnustep-gui release
Date: Fri, 16 Dec 2016 17:51:24 +0100
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Icedove/45.4.0


Le 16/12/2016 à 17:35, Ivan Vučica a écrit :
> On Fri, Dec 16, 2016 at 4:25 PM, Stefan Bidigaray
> <address@hidden <mailto:address@hidden>> wrote:
>
>     Hi Ivan, 
>     I would assume, as a minimum, the soname cannot change. GNUstep
>     does have a habit of incrementing the soname with MINOR release
>     number bumps, even if the ABI remains unchanged.
>
>     For example, Debian's current version of GNUstep base in testing
>     is 1.24.9, and the package name is libgnustep-base1.24. If a new,
>     1.25 release, is pushed out, the new package name would be
>     libgnustep-base1.25, and likely not accepted.
>
>
> Actually, the latest bump of -gui (to 0.25, the first release I was
> cutting) was actually pushed due to issues while doing Debian
> packaging. Eric *requested* that we bump the soname.
Yes, if there is ABI/API breakage, the soname must be bumped. I don't
think it is Debian specific.
>
> While searching for how to reply to this, I actually found Eric's
> off-list email, where he was running dpkg-gensymbols to generate this
> information.
>
> However, at this point, I am not sure how to go about running this
> type of thing myself. I could probably investigate, but short
> instructions or a pointer to useful documentation would be very
> time-saving.
>
> Preferably, I would be able to avoid doing a full binary package
> build, of course :)
>
> Rubber duck <https://en.wikipedia.org/wiki/Rubber_duck_debugging>
> tells me I could use 'nm' or 'readelf' or 'objdump', and then diff the
> output, but dpkg-gensymbols and its ilk might be doing much more.

I quote Yavor:

"Dpkg-gensymbols  can occasionally detect API additions or ABI breaks -- an
addition or removal of a function or of an entire class (or class
category).  However, that is completely unreliable because method
addition/removal (the most common case) is undetected."




reply via email to

[Prev in Thread] Current Thread [Next in Thread]