Naming the ARM(64) binaries for Windows

Karl Berry karl at
Thu Dec 3 00:19:14 CET 2020

    Subject: Naming the ARM(64) binaries for Windows

Does Hans have a preference?

In the abstract, I'd prefer aarch64-windows, but if Microsoft only ever
calls it "arm64", arm64-windows seems better to me. For the same reason
that we use amd64-freebsd but x86_64-linux -- that's what the users of
those particular systems are used to, so it seems nicer to follow suit
in TL than enforce a one-size-fits-all regime.

As implied already, Windows (that is, TL's "win32") is not detected by
running uname or config.guess or any such tool, since they don't
generally exist. Instead, we check Perl's builtin $^O variable (as you
can see in various places in and throughout the code). This
is one of the significant reasons why supporting two different sets of
Windows binaries in TL would be such a hassle, since presumably that
test would have to be refined. And everywhere that uses the results of
the test would also have to be inspected and possibly adapted. It is
pervasive throughout the entire infrastructure.


More information about the tex-live mailing list.