Build on Nixpkgs/NixOS #4405

jacereda · 2018-04-02T17:50:57Z

Some problems with this PR:

Still contains a crude hack (a chmod on Darwin after compiling). I think this is due to a bug in clang-wrapper, but it could well be some silly error of my own.
~~Enabled FFI+GMP by default due to my low skills with cabal and Nix.~~
~~Removes the TIMEOUT invocation for some scripts, I still need to figure out why those failed when invoked through it.~~
I didn't test anything on systems without Nixpkgs.

Maybe someone has ideas on how to deal with those.

into build-on-nixpkgs

david-christiansen · 2018-04-02T20:14:04Z

I'm not a Nix expert, so sorry if this is misguided. How do we ensure that updates to Idris don't break the Nix build script? Can we reasonably run it from Travis, for instance?

david-christiansen · 2018-04-02T20:15:06Z

Makefile


 dist/setup-config:
-	$(CABAL) configure $(CABALFLAGS)
+	$(CB) configure $(CABALFLAGS)


What is the purpose of renaming the CABAL and MAKE variables? These shorter versions are (slightly) more difficult for me to understand.

Those try to ensure the correct idris is invoked via env IDRIS=$(IDRIS) ... and I simply used those. I'm not particularly attached to any of those names though, what would you suggest?

Why not leave them as they were?

What do you mean? Repeat explicitly like in env IDRIS=$(IDRIS) $(MAKE)?

I was commenting on renaming the CABAL variable to CB here.

Uses the original names now.

david-christiansen · 2018-04-02T20:16:07Z

idris.cabal

@@ -108,12 +108,12 @@ custom-setup

 Flag FFI
  Description:  Build support for libffi
-  Default:      False


These flags should stay as they are, because libffi can be difficult for Mac users and some people don't like the license of GMP.

And I agree, I'm sort of asking how should this be handled in the second bullet point.

I think that their behavior should remain unchanged in the .cabal file at least. Nix users are a minority of Idris users, and I don't think that we should make it more difficult to build it the common way to make Nix packaging easier.

david-christiansen · 2018-04-02T20:16:58Z

idris.nix

@@ -0,0 +1,61 @@
+{ mkDerivation, aeson, annotated-wl-pprint, ansi-terminal


Is this output from cabal2nix, or is it hand-written? If it's generated code, should it be in the repo, or should the repo contain instructions to generate it? I'm not sure what best practices are around that.

It is the output from cabal2nix with some hand-edits to make sure perl, nodejs and cabal are available and to fix some problems with fsnotify (IIRC) that forgets to add some frameworks to the propagated dependencies. I'm a newbie with Nix, so I don't have an answer for the best practices.

@shlevy , I seem to recall you being an experienced Nixer. Is this the right way to do things? (sorry if I remembered wrong)

I don't think we've quite reached a point yet where there are best practices here exactly, but my preferred approach would be:

a) Autogenerate the cabal-based Nix expression on-the-fly with callCabal2nix
b) Add any necessary modifications using overrideCabal, overrideDerivation, etc. after importing the base generated expression.

This won't be helpful to you now, but we're developing a set of libraries internally at Target that we're hoping to open source over the next few months that will make this workflow clean and easy and hopefully get it adopted more widely.

I'm happy to make these changes if/when the PR is in an otherwise-acceptable state.

Thanks @shlevy! @jacereda, does the approach of importing the unmodified cabal2nix output make sense to you?

Sure, I just need to figure out how.

david-christiansen · 2018-04-02T20:17:57Z

libs/Makefile

@@ -1,37 +1,45 @@
-build:


These changes seem unrelated to the task of building Idris with nix. Could they be in a separate PR/commit? And do they maintain compatibility with BSD make?

It doesn't try to maintain compatibility with BSD make because Setup.hs already assumes GNU make (see mymake definition).

I can extract that to a separate PR if desired, but I'd prefer to keep it in this PR as well, since I had problems with missed dependencies until I changed the Makefiles.

OK. @markuspf , is this likely to cause a problem for BSD support?

david-christiansen · 2018-04-02T20:19:06Z

test/basic010/run

@@ -1,10 +1,4 @@
 #!/usr/bin/env bash
-


Why change this script? And likewise for reg039.

Third bullet point. I don't know yet why, it fails on Nixpkgs/Darwin when invoked via the TIMEOUT script.

The point of that script is to ensure that there's not an infinite loop. Getting rid of it breaks these tests, so that shouldn't be done. Better to figure out why Nix isn't working with these.

jacereda · 2018-04-02T22:04:03Z

A travis job building via nixpkgs would be the next logical step to ensure this doesn't bitrot. They already have infrastructure in place for Nix jobs.

https://docs.travis-ci.com/user/languages/nix/

david-christiansen · 2018-04-03T04:42:30Z

Good to hear that Travis supports Nix.

It looks to me like there's already an Idris package in nixpkgs: https://github.com/idris-lang/Idris-dev/wiki/Idris-on-NixOS Why can't that be used here?

david-christiansen · 2018-04-03T04:43:04Z

(and thanks for this, by the way - I'm rushing through comments and have just realized that I'm likely coming across as a bit gruff)

shlevy · 2018-04-03T11:36:31Z

@david-christiansen Re the build expression in nixpkgs and reducing duplication:

We can use the expression in nixpkgs here as long as the build process doesn't differ significantly for a release build vs a dev build. This isn't always true, but if it is for Idris then great!
We can't use the expression here in nixpkgs as of now, because there are unresolved issues with recursive nix, where a Nix evaluation depends on the results of a build (in this case, the build being the fetch of the tarball containing Idris's custom Nix expressions). In the long run we definitely want to support expressions provided by upstream packages, though.

david-christiansen · 2018-04-04T13:39:56Z

@jacereda Have you tried the expression from nixpkgs, instead of this one? Presumably it won't require all the changes to the Cabal flags and Makefiles.

david-christiansen · 2018-04-04T13:40:11Z

@shlevy Thanks!

jacereda · 2018-04-04T22:06:21Z

@david-christiansen I have successfully installed idris via nixpkgs, but I haven't figured out how to use the expressions in nixpkgs for interactive development and things like running a single test after making a change. Maybe @shlevy can help with that?

jacereda · 2018-04-05T12:16:55Z

Oh, I found there's a callCabal2nix, I guess I should use that instead of a make rule.

shlevy · 2018-04-05T13:26:49Z

idris.nix

+      export DYLD_LIBRARY_PATH=`pwd`/dist/build
+      export TARGET=`pwd`
+    '';
+    src = if lib.inNixShell then null else orig.src;


Haskell packages have an env attribute that is appropriate for nix-shell environments, have you tried that?

I'm not sure I understand. Do you mean the src = ... part is redundant?

I mean rather than doing src = ... here you can do something like let pkg = nixpkgs.pkgs.haskell.packages.${compiler}.callPackage ./idris.nix { }; in if inNixShell then pkg.env else pkg in default.nix. See the output of cabal2nix --shell for an example.

OK, I'll get rid of shell.nix and use that instead to avoid polluting the base directory.

… build-on-nixpkgs

jacereda added 7 commits March 30, 2018 20:17

Trying to enhance the experience on NixOS/Nixpkgs

ab23462

Missing space

0e0bd1a

Merge

52d9600

Merge branch 'build-on-nixpkgs' of https://github.com/jacereda/Idris-dev

44df85b

into build-on-nixpkgs

Enable tests

150e586

Merge

f723856

cabal test wasn't building the executables

8009e2b

david-christiansen reviewed Apr 2, 2018

View reviewed changes

Blind attempt at fixing Setup for Cabal-1

55dc5db

Apply stylish-haskell suggestion

af91928

jacereda added 2 commits April 4, 2018 23:46

Restore timeouts via ulimit

5b018fe

Avoid perl dependency

b6ed5d0

jacereda added 3 commits April 5, 2018 00:51

Tests failed on windows due to ulimit -t, timeout script using node

2b60976

Use another method to export IDRIS

6a2151b

Generate base nix expression with cabal2nix from Makefile

f1ef4c6

Use callCabal2nix

9db7c86

shlevy reviewed Apr 5, 2018

View reviewed changes

jacereda added 9 commits April 5, 2018 15:54

Warning

78d6af9

Patch node shebang

734e480

Get rid of shell.nix and idris.nix

caccebd

Avoid duplicated CFLAGS

4c23999

Don't force compiler

3e6e470

Use installed idris when testing on appveyor

cf3d89f

Merge master

cc13c44

Revert travis test

405c415

Leak warning

8d76d46

jacereda mentioned this pull request Apr 7, 2018

Made fromIntegerNat use case instead of if #4357

Open

jacereda added 4 commits April 19, 2018 11:16

Merge branch 'master' of https://github.com/idris-lang/Idris-dev into…

70e2514

… build-on-nixpkgs

Merge master

5e209f1

Merge master

42616f9

Merge

93d5656

		@@ -0,0 +1,61 @@
		{ mkDerivation, aeson, annotated-wl-pprint, ansi-terminal

Build on Nixpkgs/NixOS #4405

Are you sure you want to change the base?

Build on Nixpkgs/NixOS #4405

Conversation

jacereda commented Apr 2, 2018 • edited Loading

david-christiansen commented Apr 2, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jacereda commented Apr 2, 2018

david-christiansen commented Apr 3, 2018

david-christiansen commented Apr 3, 2018

shlevy commented Apr 3, 2018

david-christiansen commented Apr 4, 2018

david-christiansen commented Apr 4, 2018

jacereda commented Apr 4, 2018

jacereda commented Apr 5, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

shlevy Apr 5, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jacereda commented Apr 2, 2018 •

edited

Loading

shlevy Apr 5, 2018 •

edited

Loading