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

[Norbert Preining]

> One thing to note is that I have let install-tl configure the
> installation by importing the settings from a previous installation on
> the same disk.
> 
> I literally just started the installer, hit `y` to take over all
> settings from the TL2020 copy next to it, and then `I` to start the
> installation.
> 
> I got the message, though, when I did not import the settings, but
> started the install-tl configuration from scratch.

Ahh, good point. I will look into that. It *should* be shown also in the
case of profile/imports.

Thanks for testing!!! Much appreciated.

> > Press Enter to return to the menu, or i to install anyway.
> > ****************************************************************
> 
> I like how the lowercase `i` violates the principle of least
> astonishment (forces you to read what's happening, hard to
> predict/automate lazy workarounds).

Yes, because we know with very very high probability that it will fail.

> I tried it again, same phenomenon. I had to kill install-tl with
> SIGTERM to get my shell back.
> All HTTPS hosts that I tested hung with SSL_INIT, and install-tl is
> not interruptible:
> https://ctan.ijs.si/mirror/tlpretest/
> https://ctan.dcc.uchile.cl/tlpretest/
> https://texlive.info/tlpretest/
> 
> I then tried with http://ctan.dcc.uchile.cl/tlpretest/ because it also
> offers non-SSLed HTTP, and it works like with Utah.
> I conclude that the culprit indeed is HTTPS (and mirror issues might
> just come atop of this).

Very interesting ... very very interesting ... I need to think and
search, maybe the signal handler was overridden or something. That is
really interesting ...

> To make things more complicated, I gave the installer another chance on
> a different host, and there Ctrl-C works fine with a HTTPS mirror.

Uhmmm, then I really guess that it is bound to timing.
Ctrl-C when issued during the tar/xz run will only interrupt the tar/xz
run but not the perl run, and thus perl continues.
That means that our signal handler does not work as we expected, that
is, when sub commands are called via `..` or similar in perl, the
commands are not forwarded to perl, or if they are, we don't evaluate
them.

Again, need to read up ...

Again, thanks a lot for all your detailed experiments, that is really
great!

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