Working build

[garrison]

• The line unset NIX_ENFORCE_PURITY is, of course, reminiscent of NIX_ENFORCE_PURITY breaks ekam in hard to diagnose ways NixOS/nixpkgs#13532.
• Some test failures remained because there is no /var (and thus no /var/tmp) in the build environment. So I used sed to patch the capnproto sources in place.

This is probably not the final solution, but it at least works!

[zenhack]

Nice!

Maybe @kentonv can comment re: whether there's a better workaround for when /var/tmp isn't available?

[zenhack]

If this is going to be our solution, I think it could stand a comment explaining what the issues/with links to relevant docs.

[garrison]

Well, I can link to that previous issue (and perhaps this one), but I really have no deep understanding of what is going on so far. I figured this out by noticing that builder.sh fails under nix-shell but succeeds if the line . "$stdenv/setup" is commented out (#2). So I ran the following script under nix-shell

set > env-pre
. "$stdenv/setup"
set > env-post

... diffed the outputs, and then manually set the environment variables that were different. I bisected them (started by commenting out half of the lines) and after a few such rounds determined that NIX_ENFORCE_PURITY alone is causing the issues. Interestingly, the linked issue is pretty much the first thing that comes up when searching the internet for that identifier. I had found that issue previously, but thought it must have been closed for good reason.

Also, while the current workaround is good enough to get ekam to build, I suspect that ekam is unlikely to be able to build anything else unless the user were to unset NIX_ENFORCE_PURITY before invoking ekam in that build as well.

[garrison]

Hmm, there are a handful of other expressions in nixpkgs that either unset NIX_ENFORCE_PURITY or set it to the empty string. So maybe my solution is an acceptable one after all.

[garrison]

I find it curious that the capnproto nix expression does not require the same /var/tmp workaround. I wonder: are the tests even being run there? By nixpkgs defaults, a build that relies on autotools will at least run make check, I think. EDIT: I was wrong; see comment below.

Also, should we really be keeping the bin/capnp* files in the output build directory? Or a more basic question: what is the role of the capnproto binaries once the build is complete? It seems a bit odd to include them here, since there is nothing preventing ekam from appearing before capnproto in the PATH, and thus the binaries provided here could potentially be picked up instead of the binaries actually provided by the capnproto expression.

[garrison]

I understand this a little bit better now. The only role of NIX_ENFORCE_PURITY is to filter out paths that do not refer to the NIX_STORE (see here and here). There must be something being filtered out that ekam was expecting.

Here's the complete list of NIX_ENFORCE_PURITY's usage in nixpkgs and nix:

• https://github.com/NixOS/nixpkgs/search?q=NIX_ENFORCE_PURITY
• https://github.com/NixOS/nix/search?q=NIX_ENFORCE_PURITY

[garrison]

whether there's a better workaround for when /var/tmp isn't available

I'm not sure that it matters much, since this is only used for tests. We could, perhaps, have the tests write somewhere in the build directory tree, but I'm not sure this helps anything, as the whole build tree is (I believe) a chroot in /tmp.

[garrison]

By nixpkgs defaults, a build that relies on autotools will at least run make check, I think

I was wrong: it only happens if doCheck is set to true.