Re: Gub failure

[David Kastrup]

"Phil Holmes" <address@hidden> writes:

> ----- Original Message ----- 
> From: "David Kastrup" <address@hidden>
> To: "Werner LEMBERG" <address@hidden>
> Cc: <address@hidden>; <address@hidden>
> Sent: Sunday, July 24, 2016 5:57 PM
> Subject: Re: Gub failure
>
>
>> Werner LEMBERG <address@hidden> writes:
>>
>>>> Maybe gub still contains an older version?  Sorry for not having
>>>> checked that before committing.
>>>
>>> Indeed, I now see that only texinfo 4.11 is required.
>>>
>>> How to proceed?  Reverting is easy (commit
>>> 445bf3bb2fbd1f259fe43ade204fb34d68bdd581, the changes in
>>> `macros.itexi').  However, I would prefer if we update to a newer
>>> version of texinfo.
>>
>> Texinfo 5 is vastly slower and this might make an impact on Gub times.
>> On the other hand, stuff like the improved indexing requires fairly
>> recent versions of Texinfo.
>>
>> -- 
>> David Kastrup
>
>
> Gub takes such an extraordinary time to build, it probably wouldn't
> make much difference.  On my core i7, quad core, 8 cpu system, a full
> build from scratch takes over a day.  The shortest time I ever get
> from a good previous build is about an hour and a half.  A standard
> make doc is about 16 minutes.

Note that even "make" builds a basic non-image version of the Info
files, and of course with a non-image version, the difference is quite
noticeable.

At any rate, if we were to update Texinfo, it would likely make sense to
take the newest available version since then some of the performance
regression is likely to have seen some improvement.  I am not actually
sure about the indexing improvements, however: it may be that they are
only in PDF yet anyway, and in that case we probably already have them
since our texinfo.tex in the main tarball is much more up-to-date than
the Texinfo binaries.

-- 
David Kastrup