Re: Possible bug in 'df' v5.2.1 (--portability switch)

[Paul Eggert]

Torkel Bjørnson <address@hidden> writes:

> $ setenv BLOCK_SIZE human-readable
> $ df --portability /
> Filesystem            Size  Used Avail Use% Mounted on
> /dev/sda1             177M  111M   57M  66% /
>
> I think the last output is wrong. Since I request an output
> format in compability/POSIX mode, it should ignore the
> BLOCK_SIZE environment variable and display the blocksize
> in whatever blocksize POSIX define.

It is not a POSIX conformance bug, as the POSIX standard reserves all
environment variables like BLOCK_SIZE (see the 4th paragraph of
<http://www.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html>).
Implementations are allowed to place constraints on the values of such
variables for users who want conforming behavior.  That is why it is
OK for df to insist on POSIXLY_CORRECT for full conformance.

So this is not a standards-conformance issue.  It may desirable to
change the behavior here, but I'd like to know more about the context
first.

> Shouldn't the semantic be that when POSIXLY_CORRECT is set, this should
> override conflicting switches

No, POSIXLY_CORRECT merely causes "df" to conform to POSIX, typically
with the minimum changes to behavior.  For example, if the user
specifies both POSIXLY_CORRECT=1 and BLOCK_SIZE=human-readable, "df"
is allowed to have the latter override the former, because BLOCK_SIZE
is reserved to the implementation.