Closed HughParsonage closed 5 years ago
Authors can avoid the biber
/biblatex
runs and re-runs by just commenting out \addbibresource
and \printbibliography
. Anecdotally this results in ten-fold reduction in compile time.
Some examples using the Housing Affordability report:
One run of pdflatex
: 47s
No bibliography, no \addbibresource
: 38s
\textcite
, \footcite
disabled: 22s
graphicx
to draft: 20s
So it seems that the bottleneck really is the biblography (note that these examples are actually favourable towards bibliography speed, because biber
is not included).
The challenge is a safe and reasonable way to disable \textcite
etc.
In addition, on the question of whether it is feasible to move to bibtex
, the highest code-points by line:
char int N
1: } 125 19812
2: { 123 6748
3: ’ 8217 51
4: ~ 126 37
5: – 8211 22
6: é 233 19
7: ö 246 4
8: <U+200E> 8206 4
9: á 225 3
10: ä 228 3
11: - 8208 2
12: L 321 2
13: 160 2
14: — 8212 2
15: ó 243 2
16: … 8230 1
17: ” 8221 1
18: Å 197 1
19: è 232 1
20: ø 248 1
21: ü 252 1
char int N
i.e. about 85 / 26,719 lines have non-ASCII codes. In the new, compact bibliography there are none.
Both slower and difficult to gain speed on ShareLaTeX vs locally: 95s and 90s vs 15s and 10s locally...
No real benefit to backend=bibtex
: 16 s vs 17 s.
Turning off \CenturyFootnote
logging and reruns has minimal effect once the document is processed the first time.
General issue for performance concerns:
Issue
Possible bottlenecks
biber
/biblatex
is a known bottleneck vis-à-visbibtex
. Main issue with conversion would be stricter tests for encoding and nastier programming at thecls
file level to coerce to Grattan style.