Re: strange situation with bytes counting

[Patrice Dumas]

On Wed, Aug 21, 2013 at 12:15:18PM +0300, Sergey Poznyakoff wrote:
> Hi Patrice,
> 
> Now, info supposes that the file header (i.e. the material before the
> first ^_) is the same for each file within the split document and relies
> on this fact when computing node offsets.  

That was also my primary hypthesis.

> In the latter case the header
> length differs between split_test_before_first_node.info and
> split_test_before_first_node.info-1, that's why it is unable to find the
> Top node.  In contrast, it is able to find the chap1 node, because
> the header length in split_test_before_first_node.info-2 is the same.
> 
> So we have basically two possibilities: one is to change texi2any so
> that it treats the text between @setfilename and @top as makeinfo did.
> This will ensure compatibility with older info versions.  The other is
> to modify info so that it is able to cope with split files in new
> format.  I'm ok with either of them, but rather incline to the former
> one (because of backward compatibility).  Wdyt?

I think that we should do both.  texi2any should make sure that the same 
header is used everywhere for backward compatibility, as you suggest,
but I also think that the info readers should be able to handle
preambles of different length, in particular for the first file that
contains only tags.

In any case, the most important thing to fix is texi2any.

-- 
Pat