minutes of ESC call ...

[Stephan Bergmann]

On 11/14/2014 07:49 AM, Noel Grandin wrote:
> On 2014-11-13 06:13 PM, Michael Meeks wrote:
>> * library bloat (Michael S)
>>      + not really an issue for release builds - --enable-mergelibs
>>          + big library + --enable-symbols -> OOM
>>      + especially problematic CppuniTests with use_library_objects
>>      + perhaps a barrier to entry to new developers
>>      + Calc: (with debuginfo) 4.1 - 413Mb, 4.3 - 512Mb, master - 600Mb
>>              (stripped)                    4.3 - 26.8Mb master - 30Mb
>>          + linking in parallel gives OOM issues.
>>          => avoid making libraries even bigger.
>>          + new calc engine uses lots of templates (Markus)
>>              + mdds usage is restricted to only a few modules: (Kohei)
>>                  + 5-6 translation units only.
>>          + whole code-base is using lots more templates (Markus)
>>              + compilation speed also got more than 2x slower (Kohei)
>>                  + use of own templates & container -> STL
>>          + recently turned a lot of methods to inline methods
>> left/right (Kohei)
>
> Some suggestions:
>
> (1) Would it make a difference if I went through the *.hxx files,
> marking the inline methods with an explicit inline qualifier?

Yes, it would needlessly clutter the code.  ;)  I don't see what benefit 
you're trying to achieve here?

> (2) Perhaps we need to be using the extern template feature to limit the
> number of instantiations of various templates?

Template instantiations, via the typical "all-inline" nature of the 
template declarations, have hidden visibility and should end up with one 
copy per dynamic object.  I'm not sure whether there would be much 
compile- or run-time benefit from using another strategy (plus, I'm not 
sure whether it would work well across the various toolchains to give 
template instantiations non-hidden visibility).