Re: [NTG-context] [dev-context] integrating context mkiv, luatex, and fmtutil, mktexlsr, etc
On Mi, 03 Okt 2007, Hans Hagen wrote:
formats go to the paths defined in texmf.cnf (TEXFORMATS) and the filedatabses go alongside the ls-R files, so basically it then behaves like any other tex
??? "file databases go alongside the ls-R files" what do you mean? I thought the stuff generated by luatools is something like the ls-R database for luatex. But it is placed NOT alongside the ls-R files, right?
the cache path is only used and populated at runtime, so
So every time I run a luatex/context document it has to read ALL the trees again??? Cannot be, that was the whole point of ls-R/cache. It should be generated once (the cache).
LUATEXCACHE=$TEXMFVAR/luacache:... TEXMFSHARECACHE=yes
should be ok then; if needed, later i can look into a way to share other cache stuff (say that you generate tmc files for 300 fonts at installation time) but if i understoof right, the main reason was formats and file databases
Yes, currently the main reason is for ls-R replacement and format location. But font data could be useful, too. My system has quite a lot of fonts ... On Mi, 03 Okt 2007, Hans Hagen wrote:
texexec --make --luatex en [...] this depends on the value of TEXFORMATS nd what path is first writable
Hmm, $ export TEXFORMATS=/home/norbert/.texmf-var/web2c $ texexec --make --luatex en ... TeXExec | using tex engine luatex TeXExec | using tex format path /home/norbert/.texmf-var/web2c/luatex TeXExec | generating tex format cont-en ... Transcript written on cont-en.log. LuaTools | LuaTools | runtime: 0.18 seconds TeXExec | no lua compilations needed TeXExec | TeXExec | tex engine path: /home/norbert/.texmf-var/web2c/luatex TeXExec | TeXExec | TeXExec | runtime: 9.486118 $ ls ~/.texmf-var/web2c/luatex $ ls -l ~/luatex-cache/context/f7d1b3c25487ab1e1035aff1c53b90da/formats/ -rw-r--r-- 1 norbert norbert 5669003 2007-10-03 14:07 cont-en.fmt -rw-r--r-- 1 norbert norbert 38786 2007-10-03 14:07 cont-en.log -rw-r--r-- 1 norbert norbert 159484 2007-10-03 14:07 cont-en.lua -rw-r--r-- 1 norbert norbert 112438 2007-10-03 14:07 cont-en.luc $ So the format is placed in some strange ;-) place under luatex-cache.
btw, reminds me ... is this becoming a cross distribution linux convention (was already so on the mac) ... no longer root users, but always a sudu needed?
No, Debian does not install sudo at all per default. There is a real
root user.
-------------
To sum it up: It is a bit unclear what purpose the cache is used for:
- ls-R replacement, ie some sort of file database
- preprocessed font data cache so that loading the stuff is done faster
- formats??? what is saved for this
But all of this is somehow static. On a normal system a normal user
shouldn't have the necessity to change anything of the above. That
should be done at install time of the respective stuff.
Furthermore, the cache could be used (no idea whether this is true) for:
- single job caching of data
wouldn't it be better to keep generated files like this in the
cwd, like .aux, etc files
- caching of user files
like the above, but initiated by the user
Again, I don't see any other use than these ones, but please enlighten
me what else it is used for. But in this case some mktexlsr-ish approach
would make sense.
So suggestion, like any TEXMF tree contains a ls-R file, we could have
for any TEXMFTREE a lcache/ directory $tree/lcache/ which contains the
cache for this tree.
Since luatools and other things extract the texmf trees anyway from the
TEXMFCNF variable, the cache could be read from there.
What was the reason to do the hashing somewhere else but add the md5sums
etc for these trees?
Best wishes
Norbert
-------------------------------------------------------------------------------
Dr. Norbert Preining
Norbert Preining wrote:
On Mi, 03 Okt 2007, Hans Hagen wrote:
formats go to the paths defined in texmf.cnf (TEXFORMATS) and the filedatabses go alongside the ls-R files, so basically it then behaves like any other tex
??? "file databases go alongside the ls-R files" what do you mean? I thought the stuff generated by luatools is something like the ls-R database for luatex. But it is placed NOT alongside the ls-R files, right?
well, "TEXMFSHARECACHE=yes" makes sure that it is put in the same path as the ls-r file
the cache path is only used and populated at runtime, so
So every time I run a luatex/context document it has to read ALL the trees again??? Cannot be, that was the whole point of ls-R/cache. It should be generated once (the cache).
no, not the trees, but if one uses a font, say lmroman10-regular.otf, then the 'converterd/prepared/whatever' datatable is cached so only that file ends up in the cache (flat structure, so, when one needs lmroman10-regular.otf, i first look in the cache (no database needed) for a tmc/a file, if not present i generate one, using the otf file from the normal tree, looked up in a kpse like way)
LUATEXCACHE=$TEXMFVAR/luacache:... TEXMFSHARECACHE=yes
should be ok then; if needed, later i can look into a way to share other cache stuff (say that you generate tmc files for 300 fonts at installation time) but if i understoof right, the main reason was formats and file databases
Yes, currently the main reason is for ls-R replacement and format location. But font data could be useful, too. My system has quite a lot of fonts ...
sure, but that demands some deep thinking because currently i use version numbers in the cached data and i don't want to open all cached files with the same name in order to check
On Mi, 03 Okt 2007, Hans Hagen wrote:
texexec --make --luatex en [...] this depends on the value of TEXFORMATS nd what path is first writable
Hmm, $ export TEXFORMATS=/home/norbert/.texmf-var/web2c $ texexec --make --luatex en .... TeXExec | using tex engine luatex TeXExec | using tex format path /home/norbert/.texmf-var/web2c/luatex TeXExec | generating tex format cont-en .... Transcript written on cont-en.log.
i know -) texexec only looks there and reports what it finds; texexec itself is unaware of lua and caches and ... i need to fix that
LuaTools | LuaTools | runtime: 0.18 seconds TeXExec | no lua compilations needed TeXExec | TeXExec | tex engine path: /home/norbert/.texmf-var/web2c/luatex TeXExec | TeXExec | TeXExec | runtime: 9.486118 $ ls ~/.texmf-var/web2c/luatex $ ls -l ~/luatex-cache/context/f7d1b3c25487ab1e1035aff1c53b90da/formats/ -rw-r--r-- 1 norbert norbert 5669003 2007-10-03 14:07 cont-en.fmt -rw-r--r-- 1 norbert norbert 38786 2007-10-03 14:07 cont-en.log -rw-r--r-- 1 norbert norbert 159484 2007-10-03 14:07 cont-en.lua -rw-r--r-- 1 norbert norbert 112438 2007-10-03 14:07 cont-en.luc $
So the format is placed in some strange ;-) place under luatex-cache.
hm, also with the new version and the env var set? here it nicely goes to texmf-linux/web2c/luatex btw, the 'strange' is just an md2 of the tree
To sum it up: It is a bit unclear what purpose the cache is used for: - ls-R replacement, ie some sort of file database
indeed, more flexible, faster etc, and in the future we can use it for more
- preprocessed font data cache so that loading the stuff is done faster
indeed, only at runtime, so normally in the users path someplace
- formats??? what is saved for this
a cont-en.fmt as well as a cont-en.luc file, the format and it sstub
But all of this is somehow static. On a normal system a normal user shouldn't have the necessity to change anything of the above. That should be done at install time of the respective stuff.
no, the font cache is not pregenerated at all, think of it like this: - luatex loads otf font (takes time) - provides it as table to mkiv (takes more time) - mkiv adds a few things, preprocesses the font a bit - then saves it as table (compiled) so that loading goes in an eyeblink
Furthermore, the cache could be used (no idea whether this is true) for: - single job caching of data wouldn't it be better to keep generated files like this in the cwd, like .aux, etc files
we're talking of megabytes here btw, context has no aux, toc etc file but a tuo file where all multi pass data goes into; in mkiv there is a lua companion because much of that data is not stored in lua tables (i still have to do the table of contents)
What was the reason to do the hashing somewhere else but add the md5sums etc for these trees?
lengths of paths and such; i run luatex on the unix web/etc servers and have multiple trees in parallel, so now i can with one variable changed access all data; /temp/luatex/context/<treeroot>/whatever /temp/luatex/context/c:/blabla/tex/blabla/whatever does not work that well, so i hash the treeroot Hans ----------------------------------------------------------------- Hans Hagen | PRAGMA ADE Ridderstraat 27 | 8061 GH Hasselt | The Netherlands tel: 038 477 53 69 | fax: 038 477 53 74 | www.pragma-ade.com | www.pragma-pod.nl -----------------------------------------------------------------
HI Hans! On Mi, 03 Okt 2007, Hans Hagen wrote:
database for luatex. But it is placed NOT alongside the ls-R files, right?
well, "TEXMFSHARECACHE=yes" makes sure that it is put in the same path as the ls-r file
Ah, ok, that is with the new context you created today. No, didn't test this till now ;-) First I guess I need an updated luatex you said.
no, not the trees, but if one uses a font, say lmroman10-regular.otf, then the 'converterd/prepared/whatever' datatable is cached so only that file ends up in the cache (flat structure, so, when one needs lmroman10-regular.otf, i first look in the cache (no database needed) for a tmc/a file, if not present i generate one, using the otf file from the normal tree, looked up in a kpse like way) [...]
- preprocessed font data cache so that loading the stuff is done faster
indeed, only at runtime, so normally in the users path someplace
Hmm, I still don't get it. In all this processing, is there a job depending thing contained? I thought extracting all those tables etc can be done independently of all jobs (and will help all jobs). Anyway, I got it. So this is not a bad thing at all to have this stuff generated at job running time, so we (integrators) don't have to worry about it. Only finding all the files is important for a start, so ls-R replacement.
$ ls -l ~/luatex-cache/context/f7d1b3c25487ab1e1035aff1c53b90da/formats/ -rw-r--r-- 1 norbert norbert 5669003 2007-10-03 14:07 cont-en.fmt -rw-r--r-- 1 norbert norbert 38786 2007-10-03 14:07 cont-en.log -rw-r--r-- 1 norbert norbert 159484 2007-10-03 14:07 cont-en.lua -rw-r--r-- 1 norbert norbert 112438 2007-10-03 14:07 cont-en.luc $
So the format is placed in some strange ;-) place under luatex-cache.
hm, also with the new version and the env var set? here it nicely goes to texmf-linux/web2c/luatex
Probably an instance of "get the new beta", btw, where do I get it, from the usual download place? Or is there something else for betas?
- formats??? what is saved for this
a cont-en.fmt as well as a cont-en.luc file, the format and it sstub
Yes of course, but what needs to be cached here but the sole information that the file exist? (ie updating ls-R/cache)
But all of this is somehow static. On a normal system a normal user shouldn't have the necessity to change anything of the above. That should be done at install time of the respective stuff.
no, the font cache is not pregenerated at all, think of it like this:
- luatex loads otf font (takes time) - provides it as table to mkiv (takes more time) - mkiv adds a few things, preprocesses the font a bit - then saves it as table (compiled) so that loading goes in an eyeblink
Ok, so after the first processing later jobs will have access to these tables in a blink. Ok, don't mind. Good for me ;-) No need to worry about this.
Furthermore, the cache could be used (no idea whether this is true) for: - single job caching of data wouldn't it be better to keep generated files like this in the cwd, like .aux, etc files
we're talking of megabytes here
Are these megabytes kept and are they useful for later runs? If yes, then it is ok to put them in a separate cache (I assume the stuff is useful, it is the font stuff).
What was the reason to do the hashing somewhere else but add the md5sums etc for these trees?
lengths of paths and such; i run luatex on the unix web/etc servers and have multiple trees in parallel, so now i can with one variable changed access all data;
/temp/luatex/context/<treeroot>/whatever
/temp/luatex/context/c:/blabla/tex/blabla/whatever
does not work that well, so i hash the treeroot
Hmm, but if the caches would be in TEXMFTREE/lcache/ then the only thing
necessary would be a change of
TEXMFCNF
referencing other trees.
(Sorry, only trying to understand the details ...)
Best wishes
Norbert
-------------------------------------------------------------------------------
Dr. Norbert Preining
Norbert Preining wrote:
Are these megabytes kept and are they useful for later runs? If yes, then it is ok to put them in a separate cache (I assume the stuff is useful, it is the font stuff).
not sure yet, maybe later; but there is a cache cleaner (i may even thrash by date and usage)
Hmm, but if the caches would be in TEXMFTREE/lcache/ then the only thing necessary would be a change of TEXMFCNF referencing other trees.
(Sorry, only trying to understand the details ...)
sure, all depends on what the cache var is set too (too many variants so i play safe and hash (but i can make that piece configurable if needed) the beta is on the website and on the ftp Hans ----------------------------------------------------------------- Hans Hagen | PRAGMA ADE Ridderstraat 27 | 8061 GH Hasselt | The Netherlands tel: 038 477 53 69 | fax: 038 477 53 74 | www.pragma-ade.com | www.pragma-pod.nl -----------------------------------------------------------------
participants (2)
-
Hans Hagen -
Norbert Preining