[PATCH] Fix shared object clashes - read_data()

Uri Simchoni uri at samba.org
Fri Jan 22 09:45:24 UTC 2016

On 01/22/2016 09:36 AM, Alexander Bokovoy wrote:
> On Thu, 21 Jan 2016, Jeremy Allison wrote:
>> On Thu, Jan 21, 2016 at 08:30:42PM +0200, Uri Simchoni wrote:
>>> Well, it's not gnutls that has read_data(), it's some library
>>> (libtspi, part of trousers package) that gnutls build in FC22 pulls
>>> in.
>>> Attached is a patchset that renames read_data(), write_data(), and
>>> write_data_iov() (and fixes one bug that was found in apparently
>>> unused code).
>>> I'm for renaming because although we (samba developers) view
>>> read_data as an internal function, the linux dynamic loader doesn't
>>> seem to think so. Maybe there's some technology (linker namespaces?)
>>> that can fix this. But I'm not religious about it - the wscript-only
>>> fix seems good enough for this particular case, and adding the
>>> dependency seems justified even without the Fedora issue.
>> Where does it end ? We've been using these internal
>> names for 25+ year (I remember adding a read_data()-like
>> function in one of my first patches to Samba :-).
>> We're not using C++, we don't have a samba_ namespace.
>> Or maybe it's because I recall the trousers package has something
>> to do with an ex-Samba engineer called 'lkcl' (a blast
>> from the past, Google if you want the details) that is
>> pissing me off :-).
>> I'm happy with the wscript-only fix and I don't
>> like the rename patch, but I'm not going to be
>> an ass about it and stand in the way if others
>> disagree :-).
> Ok. Up to you. I'm fine with wscript fix.
Do I have your RB+ on the wscript fix?

More information about the samba-technical mailing list