[PyQt] PyQt releases with private SIP?

[Elvis Stansvik]

2017-02-19 2:30 GMT+01:00 Phil Thompson <phil at riverbankcomputing.com>:
> On 18 Feb 2017, at 8:31 pm, Terekhov, Mikhail <termim at gmail.com> wrote:
>>
>> On 2/18/17 13:34, Phil Thompson wrote:
>>
>>> On 18 Feb 2017, at 6:23 pm, Mikhail Terekhov <termim at gmail.com> wrote:
>>>> Hi,
>>>>
>>>> Right now every time new release of SIP and PyQt5 comes out we need to
>>>> recompile PyQt4 as well even if it was not changed. Would it be possible
>>>> to have PyQt5 release with private SIP? That would make it much easier
>>>> to have multiple PyQt versions installed simultaneously. Yes, it is
>>>> possible to have separate VE for different versions but too many VEs
>>>> quickly becomes VE hell.
>>> You only need to recompile PyQt4 if the major version of the ABI changes. That should happen very infrequently.
>>>
>> In theory that is true, but it happened several times in the past when I updated packages on OBS with new minor releases of PyQt4 and SIP but didn't update QScintilla  and then loading Qsci resulted in segmentation fault until QScintilla package was not recompiled.
>
> Yes, when the major ABI changes you have to re-compile everything (which should be very infrequently).

I'm curious here: Above, Mikhail said he updated to "[...] new *minor*
releases of PyQt4 and SIP [...]" (emphasis mine). This should imply
there were no major ABI changes, right? So his QScintilla should not
have been broken, right?

Elvis

>
>> Another case would be Windows - rebuilding here is a major undertaking and having independent versions here would be very convenient.
>>
>> BTW leaving old version forever working without maintenance and recompilation is a Good Thing (tm) :)
>
> Yes, balanced against the overhead of supporting deprecated code.
>
> The introduction of wheels has made binary (as opposed to source) compatibility much more important. It triggered the switch to the limited Python API. I acknowledge the issue you raise - but I don't think it is worth worrying about anymore.
>
> Phil
> _______________________________________________
> PyQt mailing list    PyQt at riverbankcomputing.com
> https://www.riverbankcomputing.com/mailman/listinfo/pyqt