guix: SOURCE_DATE_EPOCH is already set in some environments

[laanwj]

The environment variable SOURCE_DATE_EPOCH allows overriding the date that will be used inside the archives for guix-built binaries. This is an intentional feature, as documented in contrib/guix/README.md:

• SOURCE_DATE_EPOCH

  Override the reference UNIX timestamp used for bit-for-bit reproducibility,
  the variable name conforms to [standard][r12e/source-date-epoch].

  (defaults to the output of $(git log --format=%at -1))

However, some environments, as apparently Nix, set SOURCE_DATE_EPOCH by default. This can be extremely confusing, as it results in mismatches. i see three options:

• Show a big red warning when the Guix build scripts are called with an existing SOURCE_DATE_EPOCH. Most likely it's not what the user wanted, but they can continue if they wanted.
• Disallow overriding SOURCE_DATE_EPOCH - unset it at start.
• Rename our SOURCE_DATE_EPOCH to something non-standard that doesn't conflict with Nix.

The first one has my preference.

See bitcoin-core/guix.sigs#1201 which prompted this.

[hebasto]

• Rename our SOURCE_DATE_EPOCH to something non-standard that doesn't conflict with Nix.

As we run Guix shell in a container, it seems reasonable to rename SOURCE_DATE_EPOCH in the guix-build and guix-codesign scripts, and pass it to the container using its original name:

--env SOURCE_DATE_EPOCH="${BITCOIN_SOURCE_DATE_EPOCH:?unable to determine value}"

[laanwj]

Sounds good to me. I think it's fine to rename it, I think there's really very few edge cases in which one would really want to pass in a custom epoch to the guix build?

[fanquake]

As we run Guix shell in a container, it seems reasonable to rename SOURCE_DATE_EPOCH in the guix-build and guix-codesign scripts, and pass it to the container using its original name:

Not sure. We already have FORCE_DIRTY_WORKTREE. Seems fine to just make SOURCE_DATE_EPOCH function in the same way, rather than new variables / more build options / things being less-standard.

[laanwj]

That sounds like a good approach to me too. So, fail when SOURCE_DATE_EPOCH is set, unless FORCE_SOURCE_DATE_EPOCH is also set?

[laanwj]

Or could cover SOURCE_DATE_EPOCH being set under FORCE_DIRTY_WORKTREE, it's not strictly the same, but it's also an intentionally introduced mismatch, and would avoid adding another build option.