Added JANET_NO_AMALG flag to Makefile

[zevv]

(Not sure if people are willing to pull in this added complexity, but giving it a shot anyway)

This change allows building Janet from the individual source files instead of from the amalgamated janet.c; this considerably speeds up parallel builds on modern CPUs;

 make clean; time make -j NO_AMALG=0 
[...]
real	0m7.882s

vs

make clean; time make -j NO_AMALG=1 
[...]
real	0m1.481s

For me this make the difference between "comfortably and quickly running my tests" and "getting slightly annoyed staring at my screen",

[sogaiu]

The speedup is nice 👍

[bakpakin]

The added complexity is ok, since it is not much added complexity. We used to have this, however the requirement for distributing as janet.c and janet.h meant that sometimes the amalgamation would break, so I made the amalgamation the default in the Makefile which reduced that problem.

But a lot has changed and stabilized, I guess I just got use to waiting for the compiler 😅

[disruptek]

It would probably be sufficient to make the amalgamation a build/ci artifact, right?

[bakpakin]

Yes, but the idea was to have one fewer way to build things and I would get PRs that would break the amalgamated build (and break them myself). Basically, it was just one less thing to think about. Still, it's like a breath of fresh air to have incremental compilation nice and fast again.

[zevv]

Is the current implementation acceptable, in particular the way the option is passed to and handled in boot. janet? Also maybe change NO_AMALG into something nicer like JANET_DISABLE_AMALGAMATION? (I will not be typing that more then once (in my .bashrc only)

[sogaiu]

How about JANET_NO_AMALG?

[bakpakin]

Current implementation looks good, but it is probably better to have a name that is unlikely to collide with anything as the trigger to avoid surprises.

[zevv]

env var changed to JANET_NO_AMALG

[CosmicToast]

I'm curious on what the speed delta would look like when running with a more parallelized linker. (e.g mold)

[zevv]

Linking really is not the problem, that's typically done in tens of milliseconds for a project this size.

[sogaiu]

May be a minor thing but I think the construction (if-not (...) (do ... in the following:

    (if-not (has-value? boot/args "image-only") (do
      (print "\n/* " fname " */")
      ...))

here might be better expressed as either:

    (when (not (has-value? boot/args "image-only"))
      (print "\n/* " fname " */")
      ...)

or:

    (unless (has-value? boot/args "image-only")
      (print "\n/* " fname " */")
      ...)

The latter (using unless) is briefer, but I find it often takes me much longer to convince myself at a gut level that it's the "right thing" particularly when what's getting tested gets complicated.

(I've sometimes wanted a when-not, but I have been hesitant to suggest that kind of construction for potential bloat reasons (^^; )

---

Sorry about this remark coming after the merge -- I mentioned it at the fork earlier but I think it might not be so easy to find because of the force pushing may be?

[zevv]

Oh sorry @sogaiu, I saw your earlier remark and changed the code accordingly; this was on "my other machine" though and I forgot to push it.

[sogaiu]

Ah well, at least the code works :)