Re: On being web-friendly and why info must die

[David Kastrup <dak@gnu.org>]

Paul Eggert <eggert@cs.ucla.edu> writes:

> David Kastrup wrote:
>> I don't actually see the big deal in compilation times for a full
>> compilation.  It's not like one does it all the time.
>
> I do it all the time when I'm editing manuals, because I want to check
> the output.

But that's one manual then, not all the manuals.

> Or at least, I *used* to do it back when Texinfo was reasonably fast.

Make doc for LilyPond takes over an hour.  The time spent in Texinfo is
few minutes of that even in Texinfo 5.

Gcc was also really much slower than Turbo C in the days and I got over
it.

> Also, if I check out another branch and do a "make", the other branch
> may have different manuals and so this may force "make" to rebuild
> them, and "make" may take a long time, even when using "make -j" --
> the manuals are now *so* slow to process that they are often on the
> critical path for "make -j".

When compiling Emacs, I tend to see pretty much all of the time after
touching a large file set spent in the C compiler and
byte-compiling/autoload generating.

> I did find a simple way to avoid the performance problem: don't edit
> the documentation and don't switch branches.  But these restrictions
> are really off-putting.  We need to do better.

Well, Texinfo is a small project.  If one is serious about wanting to
invest time ("we need to do better" suggests so), it is likely spent
quite effectively there.

-- 
David Kastrup