release: 0.2.4

[elvogel]

Dependencies only. Output is byte-identical across the 13-document corpus, so nothing a
caller can observe has changed except speed.

PhoenixmlDb.Xslt 2.0.0 → 2.4.1

Taken for a measured gain rather than for currency:

2.0.0   26,609 ms
2.4.1   21,264 ms      20% faster

Min of 3, same document, rebuilt between, pin the only variable.

The release worth pausing over is the one being taken

2.4.1 is a hotfix for a native stack exhaustion, and its fix introduced a recursion-depth
ceiling that #199 — filed the same
day — says miscounts, halving effective depth for recursion that both applies a template and
constructs an element.

docmd's nested-list rebuild is exactly that shape, so it was measured:

depth  9    identical to 2.0.0
depth  50   identical to 2.0.0
depth 200   both engines stop at docmd's own serialiser guard, not the engine's

w:ilvl caps at 8, so real documents never approach the ceiling. A defect can be real, open,
and in the version you are taking without reaching your workload — and that distinction came
from testing the shape the bug describes against the constraint the inputs actually have.

Verification

Tests	361 pass, 0 failing
Output	13 of 13 corpus documents byte-identical
Locks	clean under --locked-mode
Package	icon, readme and documentation project URL intact

Not included

phoenixmldb-xslt#12 is fixed in this engine, so --style-map can drop its node-parameter
workaround — behaviour-adjacent, belongs in its own release.

The one-predicate contract stays. Chained predicates remain quadratic at 21.5 s (n=1500), so
that stylesheet comment is still accurate rather than historical.

🤖 Generated with Claude Code

https://claude.ai/code/session_018wZEgtuzaZPswykiGEBxbz