make command in bare-metal issue is wrong

[rhafer]

I suspect the docs for bare metal install are a bit misleading.

Yeah. Me too. I don't think make clean generate -C opencloud will do what the author expected it to do. (Or it depends on the concrete implementation of make. On my system (gnu make) it changes into the build dir and executes the clean, generate and build targets there. This is not what we need. We need execute the generate and clean targets in the root dir an then the build target in the opencloud dir. I guess it's safer to do that in two separate calls to make.

make clean generate
make -C opencloud build

Originally posted by @rhafer in #562

[Narrat]

The documented command did work in the past with gnu make. Maybe there was more recursion at work?
But at that time there were some changes to the Makefile which made it a bit back and forth regarding working or not.
Related:

• build failure (follow-up on #289 and #308) opencloud#309
• Make generate build as a single command #139
• correct build command #143

Wasn't without issues though:

• error when starting opencloud (r60.gb5bc6fb) opencloud#315
• fix: make recursion and missing opencloud node generate step opencloud#317

With opencloud-eu/opencloud#317 the one-liner worked.
And IIRC it failed again with opencloud-eu/opencloud#385 which is the current status.
At least I switched the instructions for the package recipes I maintain back on 2025-03-16.

Dunno what ultimatively is the more correct approach. When it was working the one-liner had its benefits, as it only compiled what was necessary for the opencloud binary. Around march the generate from the top level did considerably take longer. Something which I don't notice with the current Makefiles

[rhafer]

The documented command did work in the past with gnu make. Maybe there was more recursion at work?

It worked in a sense, that it did not exit with an error. But I am pretty sure that the generate and clean targets did not really do what they're supposed to do. AFAIK the -C arguments are evaluated before executing any of the make targets, independent of the position it appears on in the command line.

[Narrat]

-C is evaluated first, yes. And therefore I used to oder the command differently for the build recipe as I knew ordering arguments isn't a thing with make.
Cannot tell for clean as it wasn't important as every build is done from a fresh clone (and therefore omitted). But it was rather easy to detect if generate did work as intended or not. The resulting binary was around 5-6MB smaller and if run threw errors regarding the missing web. When it worked the size was comparable with the prebuilt binaries provided from the release assets.

But kinda funny that it could be a case of happenstance. I thought the Makefile may have checked on pre-conditions but back then I didn't look that deep into the Makefiles or the inner workings.