[hackers] Re: [stali][patch] Change dependency specifications in makefiles

From: Evan Gates <evan.gates_AT_gmail.com>
Date: Wed, 7 Sep 2016 17:34:15 -0700

I forgot to add, one important side change is that after updating a
library, all packages that depend on it will now automatically rebuild due
to moving library dependencies out of LDFLAGS and into LIBS and specifying
that the BIN depends on LIBS.

On Sep 7, 2016 4:29 PM, "Evan Gates" <evan.gates_AT_gmail.com> wrote:

> I've made a lot of changes to the makefiles to specify dependencies
> correctly and allow parallel builds with make -j. Please be on the
> lookout for bugs and please bring up any objections/criticism.
>
> Specify dependencies correctly to allow parallel builds with -j.
> Previously the makefiles relied upon sequential execution in order to
> ensure prerequisites were built first. The new dependencies are done
> on a per package level and a per directory level. For example parts of
> the parted package depend on parts of the e2fsprogs package. Instead
> of listing the exact dependencies and recipes in the parted makefile,
> list e2fsprog as a prerequisite for parted in the bin/stali.mk. While
> this isn't as exact, and means that you can't build parted directly
> from its directory without building anything else, it sucks much less
> to implement and puts dependencies in a more central location and
> makes them easier to read and add to.
>
> Using guard files to protect against redundant recipe execution. In
> some instances more than one target is built by a single execution of
> a recipe. For example, the awk package uses yacc to create both ytab.h
> and ytab.c from awkgram.y. However when doing a parallel build the
> recipe would be executed twice in parallel, once to create ytab.h and
> once to create ytab.c. In order to avoid this an intermediate rule is
> used. The rule could be PHONY, but then it would run every time
> instead of only when the grammar is updated. Instead the intermediate
> rule, called _ytab, touches a file called _ytab so we have a
> modification time to compare against. I used the convention of guard
> files begining with an underscore.
>
Received on Thu Sep 08 2016 - 02:34:15 CEST