Re: [PATCH] Documentation: fix build with Asciidoctor 2

On 2019-09-08 at 21:21:22, Jeff King wrote:

Trying again, I'm able to reproduce this. I found the cause, which is in the stylesheets. XSLT stylesheets have the ability to specify elements from which whitespace should be stripped (using the xsl:strip-space directive). In the DocBook stylesheets, listitem is specified as such an element, so the whitespace there should be stripped.

However, in DocBook 5, our elements are in a namespace. Therefore, the unnamespaced stylesheets specify only "listitem", not "d:listitem", like the namespaced stylesheets do. Because this happens right after the tree has been constructed "but before it is otherwise processed by XSLT" and isn't affected by the EXSLT extension that allows re-parsing the modified tree, then we end up with the whitespace that we don't want.

There are a couple ways around this.

1. We can force xmlto to use the DocBook 5 stylesheets with the "-x"
flag, but we have to know where they are.  Debian and Fedora have them
in different places, so we'd need a knob to figure out where they are.

2. We can force xmlto to use a custom stylesheet with "-x" that merely
imports the DocBook 5 stylesheets using a URL.  If the user has the
DocBook 5 stylesheets installed and XML catalogs configured (the default
on Linux distributions), then everything will just work and the system
will resolve it to the local copy.  If, however, things are not properly
configured, this will result in multiple network downloads for each
manual page.

3. We can give up on xmlto and do things ourselves.  This has the same
problem as option 1, since we need to learn how to find the stylesheets.

4. We can send a patch to xmlto to make it use the proper stylesheets
for DocBook 5 and hope upstream does a release and everyone upgrades
shortly.  Since xmlto is not at all active upstream, this seems like it
may be an imprudent choice.

5. We can send a patch to the DocBook stylesheets and have them include
both the namespaced and unnamespaced versions of the element names in
both sets of stylesheets and hope everyone upgrades.

My personal preference is #2; I think that seems like the best choice forward. XML catalogs are well understood and well configured on Linux distributions. Homebrew supports them adequately, but you have to add an environment variable to your shell configuration to enable them. Of course, if you're doing _anything_ with XML, you'll have them enabled.