Numbers

[Rob Beezer]

Announcement over on pretext-announce about refactoring of numbering code. A
few notes for interested developers.

We now do a depth-first traversal of the document tree in the assembly
(pre-processor) phase, checking each structural division's level versus the
desired level of numbering given in the publisher file. At "terminal" divisions
(leaves) we form a node-set of all items to number, and compute an object's
serial number from its location in that node-set. Once we descend below the
numbering level, we form a node-set of all the items to number in the whole
subtree below a division and compute serial numbers relative to that collection.
Recorded as attributes, they are trivial to use when needed (rather than being
re-computed!). All very natural in XSL. (Skipping a few fine points!)

Largely this matches what LaTeX does with standard/rational setups for counters.

A new set of commits will add numbering for "open problem" blocks, partly as a
tutorial on how to support new blocks (which I was afraid to do before).

The sample article now builds as HTML for me in 4 seconds, rather than 5
seconds. The XSL itself is about 25% faster (thanks Andrew for the profiling
tools!). This is due in part to the sample article having extensive examples of
#xref (which would recompute numbers before). But other examples (e.g. sample
book), show gains of 10-15% or so. So some real improvement, and lots of old
bad code is gone.

We now have (optional) separate (distinct) counters for blocks: figure-like,
project-like, inline exercise, and soon for "open problems". And rational
publisher entries to match, plus documentation. I feel like that is enough.

There is now a "numbering test document." This was written by Claude AI, for
use by Claude AI. But it is worth knowing about - lots of structure, lots of
numbered objects (many in oddball places), and just enough context so Claude
does not get lost (I do, it is a mess!). Recent changes sometimes involved over
100 builds to check before/after, and HTML/PDF differences across many varied
sets of publisher options for numbering.

The numbering code got moved around to different stylesheets and an xsl:import
is now a xsl:include. That all seems fine, but problems there are sometimes
hard to see in advance. Claude was a lot of help with this, organizing a
sequnce of commits that allowed for careful testing at each step.

Holler if you see any odd behavior, and we'll keep an eye on pretext-support.

Rob