To speed up the initial build, I tried to replace big collections with a minimal set of specific packages, leaving texliveonfly to do the hard work of installing more.
Performance
I just did a fresh deploy before and after this PR on Netlify (with cache cleared), and got a 1m 58s = 58% reduction. The improvement is more impressive if you just measure the texlive installation time (time between [build.sh] Installing TeX Live... and [build.sh] Installed TeX Live. messages): almost 3x faster!
Version
TeX Live install
texliveonfly run
Total build (as billed by Netlify)
Before PR
2m 22s
0m 49s
3m 23s
Alternative
0m 48s
1m 24s
2m 25s
After PR
0m 43s
0m 32s
1m 25s
Of course, there is probably some measurement error here, and the improvement might get better or worse for different examples. But this PR seems in the spirit of this build script of "install just enough and let texliveonfly deal with the rest". And given that it improves both the TeX Live install time and the texliveonfly run, it seems like a pretty sure win. (See below for description of the "Alternative" row in the middle, which was a first version of this PR, where the winning was less clear, where texliveonfly had to do a lot more work.)
Description: "These files are regarded as basic for any TeX system, covering longdesc plain TeX macros, Computer Modern fonts, and configuration for longdesc common drivers; no LaTeX."
This seems fairly important, so I've left it in: it has some extra unnecessary items like xdvi, but probably hard to tease apart.
collection-binextra (TeX auxiliary programs)
Description: "Myriad additional TeX-related support programs. Includes programs and macros for DVI file manipulation, literate programming, patgen, and plenty more."
This is a pretty small set, and actually do seem like pretty common. In the first version of this PR ("Alternative" row above), I excluded this and just installed latex latex-bin. But this caused texliveonfly to do a lot more work. So in the final version of the PR, I kept this in.
(Item 1 from https://github.com/frangio/netlify-latex/issues/6#issuecomment-707937166)
To speed up the initial build, I tried to replace big collections with a minimal set of specific packages, leaving texliveonfly to do the hard work of installing more.
Performance
I just did a fresh deploy before and after this PR on Netlify (with cache cleared), and got a 1m 58s = 58% reduction. The improvement is more impressive if you just measure the texlive installation time (time between
[build.sh] Installing TeX Live...and[build.sh] Installed TeX Live.messages): almost 3x faster!Of course, there is probably some measurement error here, and the improvement might get better or worse for different examples. But this PR seems in the spirit of this build script of "install just enough and let texliveonfly deal with the rest". And given that it improves both the TeX Live install time and the texliveonfly run, it seems like a pretty sure win. (See below for description of the "Alternative" row in the middle, which was a first version of this PR, where the winning was less clear, where texliveonfly had to do a lot more work.)
Which collections to include?
http://mirror.ctan.org/systems/texlive/tlnet/tlpkg/texlive.tlpdb lists the contents of the various texlive collections. Here are the details for the three that were installed:
amsfonts bibtex cm colorprofiles dvipdfmx dvips ec enctex etex etex-pkg glyphlist graphics-def hyph-utf8 hyphen-base hyphenex ifplatform iftex knuth-lib knuth-local kpathsea lua-alt-getopt luahbtex luatex makeindex metafont mflogo mfware modes pdftex plain tex tex-ini-files texlive-common texlive-docindex texlive-en texlive-msg-translations texlive-scripts texlive.infra tlshell unicode-data xdvixdvi, but probably hard to tease apart.a2ping adhocfilelist arara asymptote bibtex8 bibtexu bundledoc checklistings chklref chktex clojure-pamphlet cluttex collection-basic ctan-o-mat ctan_chk ctanbib ctanify ctanupload ctie cweb de-macro detex dtl dtxgen dvi2tty dviasm dvicopy dvidvi dviinfox dviljk dviout-util dvipng dvipos dvisvgm findhyph fragmaster git-latexdiff gsftopk hook-pre-commit-pkg installfont ketcindy lacheck latex-git-log latex-papersize latex2man latex2nemeth latexdiff latexfileversion latexindent latexmk latexpand light-latex-make listings-ext ltxfileinfo ltximg luajittex make4ht match_parens mflua mkjobtexmf patgen pdfbook2 pdfcrop pdfjam pdflatexpicscale pdftex-quiet pdftosrc pdfxup pfarrei pkfix pkfix-helper purifyeps pythontex seetexk spix srcredact sty2dtx synctex tex4ebook texcount texdef texdiff texdirflatten texdoc texdoctk texfot texlive-scripts-extra texliveonfly texloganalyser texosquery texplate texware tie tlcockpit tpic2pdftex typeoutfileinfo web xindex xindy xpdfopentexliveonflyandlatexmk, so just install those instead.ae amscls amsmath atbegshi atveryend auxhook babel babel-english babelbib bigintcalc bitset bookmark carlisle collection-basic colortbl epstopdf-pkg etexcmds fancyhdr firstaid fix2col geometry gettitlestring graphics graphics-cfg grfext hopatch hycolor hyperref intcalc kvdefinekeys kvoptions kvsetkeys l3backend l3kernel latex latex-bin latex-fonts latexconfig letltxmacro ltxcmds ltxmisc mfnfss mptopdf natbib oberdiek pagesel pdfescape pslatex psnfss pspicture refcount rerunfilecheck stringenc tools uniquecounter urllatex latex-bin. But this causedtexliveonflyto do a lot more work. So in the final version of the PR, I kept this in.