[tlbuild] tlmgr/install-tl bug: running out of disk space

Norbert Preining norbert at preining.info
Sun Mar 13 03:24:39 CET 2022


Hi Johannes,

thanks for testing!!

> 1. I like, in principle, that install-tl is complaining that disk space
> is short:

You used the text installer, right? With that you should have seen a
HUGE warning just above this line:
> 
> Enter command: I

*** WARNING ****************************************************
The installation requires $vars{'total_size'}M of disk space
but only $vars{'free_size'}M is available.

You probably want to either clean up the destination filesystem,
or choose a different installation location,
or reduce what gets installed.

Press Enter to return to the menu, or i to install anyway.
****************************************************************

> Installing [1/4, time/total: ??:??/??:??]: hyphen-base [22k]
> Insufficient disk space, but continuing anyway.Installing [2/4, time/total: 00:00/00:00]: kpathsea [1087k]

The warning here is going to STDERR and thus is shown so strange.
I will simply disable it, I guess there is no need for it.

> 2. It depended on the mirror if Ctrl-C works or not.
> >From https://ctan.ijs.si/mirror/tlpretest/ , Ctrl-C does NOT abort

That is **absolutely** unbelievable, since there is no single line
dependent on the actual mirror. I guess it is a timing issue, that is,
the Ctrl-C is received while the tar is running or while the perl code
is running. I thought the signal handler would deal with that ... though
...

> 1. There might be an issue with the ijs.si servers and/or the network
> between there and me (access via web browser intermittently didn't work
> smoothly for some time either), or a DDoS protection.

Nothing I/we can do about it.

> 2. HTTP vs. HTTPS? The Utah site is plain HTTP, while Ctrl-C with IJS
> hung at SSL_INIT.

No idea ..

> When the disk ran out of space again, install-tl throws the same error

Yes, that is as expected. You got above big fat warning and agreed to
continue. We don't do anything more, mostly because *actual* free space
and the one reported by "df" might not agree (COW file systems).

> I think we can live with the current behaviour. If someone's disk runs
> out of space, there is really not much to do from the install-tl side,
> and the user should sort out other issues first.

And read the warning ....

Best

Norbert

--
PREINING Norbert                              https://www.preining.info
Fujitsu Research     +    IFMGA Guide     +    TU Wien    +    TeX Live
GPG: 0x860CDC13   fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13


More information about the tlbuild mailing list.