Tested on linux.

With this change applied the output of make -d in regards to target 'all' is the following.

Considering target file 'all'.
File 'all' does not exist.
Looking for an implicit rule for 'all'.
No implicit rule found for 'all'.

Because this build system does not use make builtin rules or suffixes

What about the dependencies?

What about the dependencies?

Indeed. This wont work for at least miniupnpc:

Building miniupnpc...
make[1]: Entering directory '/tmp/cirrus-ci-build/depends/work/build/x86_64-apple-darwin18/miniupnpc/2.2.2-19f39f7d911'
/bin/sh updateminiupnpcstrings.sh
make[1]: *** No rule to make target 'addr_is_reserved.o', needed by 'libminiupnpc.a'.  Stop.
make[1]: *** Waiting for unfinished jobs....

Note that there is also some related discussion in #18821.

Indeed.
Will add unexport.

One thing to consider about unexport of MAKEFLAGS is that it will prevent any flag to be propagated.
We'll need to have something like the following

saved_makeflags:=$(MAKEFLAGS)
MAKEFLAGS+=-r
unexport MAKEFLAGS

$($(1)_built): | $($(1)_configured)
    $(AT)echo Building $(1)...
    $(AT)mkdir -p $$(@D)
    $(AT)+cd $$(@D); MAKEFLAGS=$(saved_makeflags) $($(1)_build_env) $(call $(1)_build_cmds, $(1))
    $(AT)touch $$@

This should let the options specified by the user on the command line to propagate.
Let me test this.

Is this related to #22134?

Sorry, but I'm not convinced this is a worthwhile change. It adds extra settings for a very small gain. Can you at least quantify the performance win? I would not expect much of the build time to be spent in make itself.

Sorry, but I'm not convinced this is a worthwhile change. It adds extra settings for a very small gain. Can you at least quantify the performance win? I would not expect much of the build time to be spent in make itself.

I think the potential win here is not w/re performance, but w/re explicitness. I remember struggling with builtin variables and target definitions while working on 3d6603e, and I think being explicit about where we expect to use the builtin vars and targets is a good thing (also for debugging purposes) if it can be done right.

Performance wise -r helps when there is nothing to be built and your program needs to link a lot of libraries. -r helps each library a little bit.
For depends subsystem, -r won't affect performance much, especially given #22134.

-r helps troubleshooting a lot.

E.g. this is the difference in debugging output.

$ ls -l depends/packages/*.mk |wc
     29     261    2398
$ cat d.mk
include depends/packages/*.mk
all:
$ make -d -f d.mk |wc
  12752   64905  758491
$ make -rd -f d.mk |wc
    197    1200   12678

It has been my experience that people get buried in this output and do not realize they can drastically simplify it.

Is this related to #22134?

Not related.

The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.

Conflicts

Reviewers, this pull request conflicts with the following ones:

• #24285 (build, refactor: Drop useless call Make function by hebasto)
• #24279 (build: Make $(package)_*_env available to all $(package)_*_cmds by hebasto)
• #22811 (build: Fix depends build system when working with subtargets by hebasto)
• #19952 (build, ci: Add file-based logging for individual packages by hebasto)

If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.

GNU Make says make will provide information necessary for accessing the jobserver through the environment to its children, in the MAKEFLAGS environment variable.

Won't this break that?

Options specified by the user on the command line are all exported to make's children.
Including -j and --jobserver-auth.
This change only prevents the -r flag, set in this very makefile, from being exported.
If the user specifies -r on the command line, then -r is exported.

Concept ACK

• Our depends/ system in itself (not including the sub-makes it calls) does not (and likely should not) use any of the built-in rules/definitions, and the current changeset limits the effect of the -r to only depends itself and not its sub-makes (this is why miniupnpc still builds)
• All UNIX make implementations (GNU and BSD) support the -r flag (apparently there is no busybox/alpine make)

@dgoncharov Would be nice to include some illustrative examples in your comments in depends/Makefile

i added a test package here https://github.com/dgoncharov/bitcoin/tree/test_package.
If you checkout this branch, you can see the value of MAKEFLAGS, that make passes to its children.
E.g.

$ TEST=1 make test_configured -k -j2 --no-print-directory 
Extracting test...
Preprocessing test...
Configuring test...
in print_makeflags.mk MAKEFLAGS=k --no-print-directory
in print_makeflags.mk in the recipe MAKEFLAGS=k -j2 --jobserver-auth=4,5 --no-print-directory
make[1]: 'all' is up to date.
$

You can then pass various flags and see the output.
Or whatever other test you want to run.
There is simply too many combinations to list them in the makefile itself.

🐙 This pull request conflicts with the target branch and needs rebase.

_{Want to unsubscribe from rebase notifications on this pull request? Just convert this pull request to a "draft".}

Closing this as it has not had any activity in a while. If you are interested in continuing work on this, please leave a comment so that it can be reopened.

cc @dongcarl

Picked up in #33045.