build/makefile improvements

[2mol]

So this is a first baby step to make the build process do a little less unnecessary work. For now it only makes sure that for the single build process everything is compiled with the same resolver.

tl;dr: it makes it download one less stack resolver and build about a 30% faster.

This is not complete, and likely makes the building of all 8 versions slightly worse, since cabal might be builtg 8 times. (AFAIK otherwise make is smart enough to build cabal only once, even if we loop over hie-% 8 times.)

Test

I ran the builds on a macbook air i5.

Prep

Clean up everything:

rm -rf .stack-work ~/.stack ~/.local/bin/hie* ~/.local/bin/happy ~/.local/bin/cabal

Single HIE build

Tested with 8.4.4, including docs.

make hie-8.4.4 && make build-doc-8.4.4 && date

Makefile change	compilation time	~/.stack size	~/.local/bin size
unchanged	1h	5.3G (2 resolvers)	334M (3 x hie + cabal & happy)
only use one resolver	40min	3.5G (1 resolver)	334M
without building cabal
linking instead of copying

Build & install all HIE (currently 8 versions)

TODO

[2mol]

Tested on Linux, and there I still have the very weird problem that two resolvers are being downloaded.

I can't figure out right now where the command comes from that triggers the downloading of the nightly resolver.

It happens in between installing hie and generating the docs.

...
Copied executables to /home/bla/.local/bin:
- hie
- hie-wrapper
Downloading nightly-2018-12-27 build plan ...
Downloaded nightly-2018-12-27 build plan.
Preparing to install GHC (tinfo6) to an isolated location.
This will not interfere with any system-level installation.
Preparing to download ghc-tinfo6-8.6.3 ...
ghc-tinfo6-8.6.3: download has begun
ghc-tinfo6-8.6.3:    7.49 MiB / 173.91 MiB (  4.30%) downloaded...
ghc-tinfo6-8.6.3:   35.69 MiB / 173.91 MiB ( 20.52%) downloaded...
ghc-tinfo6-8.6.3:   64.36 MiB / 173.91 MiB ( 37.01%) downloaded...
ghc-tinfo6-8.6.3:   92.71 MiB / 173.91 MiB ( 53.31%) downloaded...
ghc-tinfo6-8.6.3:  121.64 MiB / 173.91 MiB ( 69.95%) downloaded...
ghc-tinfo6-8.6.3:  150.83 MiB / 173.91 MiB ( 86.73%) downloaded...
ghc-tinfo6-8.6.3:  173.91 MiB / 173.91 MiB (100.00%) downloaded...
Downloaded ghc-tinfo6-8.6.3.
Unpacking GHC into /home/bla/.stack/programs/x86_64-linux/ghc-tinfo6-8.6.3.temp/ ...
Configuring GHC ...
Installing GHC ...
Installed GHC.
stack --stack-yaml=stack-8.4.4.yaml exec hoogle generate
Starting generate
Reading Cabal... 6.25s
...

[alanz]

That looks like something is running against the default stack.yaml, which now uses a recent nightly.

So look for a stack command without a --stack-yaml= in it.

Or, rename stack.yaml and see what breaks. Except it may just use the default (global) project.

[alanz]

We do the build before install because it should be a no-op, but all 'stack install' checks is that the artifact exists, so will not rebuild changes if this is an updated workspace.

[alanz]

LGTM

To me it makes sense to merge this, since we are still deciding on using the install.hs script only.

[2mol]

let me check first that it doesn't make the build-all case significantly worse :)

[fendor]

I think, this PR can be closed since there is no makefile anymore.