Re: Flood of commits from July?

[Patrice Dumas]

On Sun, Oct 06, 2024 at 12:09:45PM +0100, Gavin Smith wrote:
> On Sun, Oct 06, 2024 at 09:53:22AM +0200, Patrice Dumas wrote:
> > > How much slower would the linear search actually be?
> > 
> > It is much slower (if I recall well, it was something like 100 times
> > slower for the texi2any manual).  I juste tested the overall effect and
> > for the pure C demonstrator for the Texinfo manual, time is about 0.375
> > s with linear search and 0.305 with the hash map.

> Does it scale with the size of the input?  Perhaps a certain part of
> this 0.07 sec is fixed regardless of the size of the input file.  Have
> you tried it with a larger manual like glibc or elisp?

I think that it should be worse for bigger manuals, as the linear search
time increases with the number of entries for each search, while the
hashmap time traversal should be less dependent to the number of
entries.

For the elisp manual, the times are 1.12s with hashmaps and 1.51s
with linear search.

It should also depends on the number of cross references in the manual.

> > After more thinking, I wonder if it wouldn't be better to require
> > setting a configure flag to have the teximakehtml demonstrator program
> > built instead of moving the C++ code to contrib, as it would be much
> > simpler.
> 
> The configure flag would be a good thing as at present, object files
> with names like ./parsetexi/teximakehtml-context_stack.o are created
> throughout the build directory.

I do not see that being an issue?  There is a need to rebuild everything
with different CPP flags, for files linked against Perl or not.

Even if we do not ship the program, checking that it builds seems to me
a good thing as it tests the C interface.

> I will try to understand how the C++ source file is used to see if I
> can propose any way of separating it better.

Ok.  I will add a configure flag too.

-- 
Pat