Re: Sizes of TL distribution

Hi Max

> On Sat, 2026-01-17 at 18:19 +0100, Frank Mittelbach wrote:
>> Now one can argue that this is a user choice and equiv of having the
>> documentation and then deleting it, but it isn't really, if it is a
>> recommendation.
> 
> The default setting is to install the documentation; users need to
> manually and explicitly choose to not install it. The "recommendation"
> here was just my personal recommendation, nothing official in any way.

sure, but ... when it gets baked into guidelines by Peter or others it 
becomes a recommendation of some sort

[...]

> Sure you can:
> 
>      https://github.com/TeX-Live/texdoc/pull/79
> 
> Testing this on a fresh installation gives me
> 
>      $ texdoc source2e
>      You don't appear to have any local documentation installed.
> 
>      There may be online documentation available for "source2e" at
>          https://texdoc.org/serve/source2e/0
>      This documentation may be for a different version than you have installed.
> 
>      Would you like to search online? (y/N)

true, but that would be even better if it would offer to set things up 
to always search online of documentation is not found
> 
> and pressing "y" opens my browser up to a copy of source2e.pdf. Now,
> this won't work offline, but considering that most users just use
> Overleaf, 

as an aside: I think that statement about most people use ...  is not 
true (and I'm not counting the few people that haven't heard of 
Overleaf) but while the numbers from Overleaf are truly impressive it is 
still only a fraction of users overall. And many are from countries 
where overleaf has probably little presence

but yes with a good internet and a decent texdoc online interface my 
argument becomes less viable (and perhaps just shows my age)


>> I think the biggest problem is that CTAN hsots far too many stuff
>> accumulated over the years.
> 
> Ok, so are you volunteering to tell CTAN which packages they should
> delete? :)

I already volunteer for far too much stuff.

> But realistically, I don't think that we could possibly
> remove anything that's been updated in the past 10 years, but all the
> old packages from the 90s have tiny file sizes, so I don't think that
> removing obsolete packages from CTAN would make a meaningful impact on
> installation sizes.

unchanged doesn't mean unused so one would have to be careful about 
that. It might only be a few packages but 10 years is not a good indicator.

> The biggest issues are fonts and random university thesis classes, but
> people use all of those, so we can't get rid of them.

yes that is kind of the problem but I think only because we need so much 
font space for pdfTeX (while with luatex or other systems like Typst I 
guess the fonts don't count as they are whatever is on the system)

So you could gain much but having a starter scheme that only provide a 
100+ font families for use with pdfTeX and not everything under the sun 
converted to Type1. That what I tried with TLC3: have a good selection 
as recommended as high quality fonts to me by a reputable font designer.

And you could make that nicely optional.

>> I think it should be possible to make a
>> medium scheme that doing the trick,
> 
> The schemes are essentially just random ad-hoc collections of packages
> (Karl, please correct me if I'm wrong here), but after writing TLC3,
> scraping arXiv for the accessibility package status table, and dealing
> with various user support requests over the years, you probably have a
> much better idea of which packages are commonly used than we do. So if
> you're willing to give us a list of packages, we'll gladly update one of
> the schemes or create a new one to contain only those packages.

it depends a bit on the target audience for the scheme, but yes 
something like arXiv gives a good indication of what is needed in STEM 
publishing as long as it is largely English based and I would claim TLC3 
also gives up a good starter scheme. I'm currently waiting of good data 
from Overleaf and all three together should make up a reasonable 
collection (but of course not if you are interested in specialized 
fields also served by TeX in one way or another ... then you are soon at 
the full scheme).

the table at

    https://latex3.github.io/tagging-project/tagging-status/full

might be a good resource. You can sort it by "ArXiv %" which gives you 
all packages used in the archive down to 0.01% (but that is still a 
sizable number on such a big archive) On the other hand, is it really 
helpful to carry around even and everything if  there are many choices 
for achieving something and and few hundred papers have made use of it?

Then there is TLC3 for which I went through the whole of CTAN at the 
time and (skim)read every documentation there and out of whichTLC3 
became my recommendation (with choices) to cover a wide variety of 
documentation tasks. That too can be used as a selection in the table.
And I hope to get similar data from Overleaf into that mix.

> But at the end of the day, we'll still have the problem of users needing
> random university thesis classes, and since very few package authors
> provide dependency information, users can't just run "tlmgr install
> university-thesis"; they'll instead need to trace down the entire
> dependency chain and install everything manually.

true, but the question is how much of that is not covered by the listI 
just mentioned? Other than the university-thesis class itself?
> 
>> and 3gb is not that bad. And even 9
>> is not really that much of a problem these days. I have a lot of
>> software that take up several gbs.
> 
> Sure, if you're committed to using LaTeX, it's really not much of an
> issue---most of us on this list have multiple versions of TL installed,

no, I meant several software products I use require more than a few MB 
but have GB sizes


> totalling to a few dozen gigabytes. But if a new user is deciding on
> which typesetting system to use for the first time, then Overleaf takes
> 0 MB of disk space,

sure, you can't compete with that number, except that you have to pay 
for it, because free is rather limited these days (or you have to have a 
university account).

> Typst takes <100 MB

you can't compete with that either, except a) it probably doesn't have 
fonts and b) what does it give your for the 100 MB?

You could probably make a small distribution if the only engine you 
install is luatex  and you use mainly system fonts (or put an OpenType 
font scheme next to it as a separate option) that will then probably get 
you in the ball park of ConTeXt.

> , MiKTeX takes ~500 MB, and the

yes but this is due to it updating and loading on the fly --- something 
that would be nice to have for TL too (at least in a restricted way)

> ConTeXt Standalone Distribution takes ~600 MB. To a new user, all these
> options appear equivalent, so then why would they want to pick the 9 GB
> option?

if you take those numbers as the basis without any other considerations 
(and people new to the topic may not have any good ways to make a 
judgement call on anything other than money or size (or make America 
great again)) then neither 9gb or 4gb or even 2gb would make the cut, 
would it?

So if you want to compete on that level you may effectively have to 
think about support for some sort of on-the-fly update mechanism (with 
all the possible caveats).

'night
frank