Web lists-archives.com

Re: Autodetection of packages that need visibility annotations

On Tue, May 15, 2018 at 10:56 AM, Simon McVittie <smcv@xxxxxxxxxx> wrote:
> On Tue, 15 May 2018 at 10:44:27 +0100, Yuri Gribov wrote:
>> On Tue, May 15, 2018 at 9:22 AM, Andrey Rahmatullin <wrar@xxxxxxxxxx> wrote:
>> > There is an interesting question about this: technically speaking,
>> > removing exported symbols requires a soname bump. But one can say that
>> > those symbols were never public and so they are not a part of ABI.
>> > Fixing this upstream and bumping the soname should be easier rthough..
>> Right, TBH until now I haven't bothered with soname bumps when
>> submitting visibility-related patches to upstream...
> If the symbols aren't considered to be public (the upstream developer
> should know) then you shouldn't bump the SONAME, to avoid unnecessary
> ABI transitions and disruption.

As a side note, in one of my other projects I tried scanning reverse
package dependencies (recursively) to detect various symbol anomalies
but that required amounts of storage and compute power well beyond of
what a single workstation can provide...