Re: 4.4: embeds volatile information in MAKEFLAGS

[Paul Smith]

On Mon, 2022-11-07 at 20:32 +0100, Steffen Nurpmeso wrote:
> mkdir /tmp/t
> cd /tmp/t
> cat > makefile <<'EOT'
> .NOTPARALLEL:
> .WAIT: # Luckily BSD make supports specifying this as target, too
> config:
>         @{ echo MAKE=\"$(MAKE)\";echo MAKEFLAGS=\"$(MAKEFLAGS)\"; } >
> ./au;echo config
> all:
>         . ./au;echo all;$${MAKE} internal_
> internal_:
>         @echo MAKE=$(MAKE), MAKEFLAGS=$(MAKEFLAGS)
> EOT

I see.  I don't think any of us envisioned this idea of storing the
MAKEFLAGS from a previous run and using them in a subsequent run.  That
is an interesting use-case.

> I mean, _if_ the FIFO does not exist then "the server exited
> gracefully", which it would not do upon crash, no?  Just ignore
> ENOENT thus?  What do you say?

I agree that we should drop down to "jobserver unavailable" here rather
than having a fatal error.

> Btw if it would be me, i would pass that via environment
> variables, which makes a natural connection to a server and its
> child clients.  I mean, Linux and GNU LibC now make (some aspects
> of) environment inaccessible for root aka SUID i think (E?UID
> mismatches, issetuid or what-do-i-know, but this hardly is
> a problem for GNU make.

I'm not sure I understand what you are saying.  These values ARE passed
to the child clients through the environment.  That's what MAKEFLAGS
_is_, it's the environment variable that's used to allow a parent make
to provide information on options etc. to its sub-makes.

The POSIX spec says (in the Rationale):

> The environment [of the child processes] is the same as the
> environment to make except that MAKEFLAGS [...] are added