rocksdb: Build with MinGW #73316
rocksdb: Build with MinGW #73316
Conversation
|
https://github.com/facebook/rocksdb/pull/6027/files almost ready! |
| ; | ||
|
|
||
| doCheck = false; # tests take a very long time | ||
| checkTarget = "test"; | ||
|
|
||
| # TODO remove | ||
| postInstall = stdenv.lib.optionalString (!enableStatic) "rm $out/lib/*.a"; | ||
| # TODO(@Ericson2314): Make resusable setup hook for this issue on Windows. |
There was a problem hiding this comment.
It looks like lz4 intentionally puts it in bindir. Perhaps the setup hook should instead add $out/bin to the -L lookups:
https://github.com/lz4/lz4/blob/49c20c498b80e70bae3ae604ef468921d79d2121/lib/Makefile#L182-L187
There was a problem hiding this comment.
This is something slightly awkward about windows, shared lirbaries (should) end up in bin. There are a few cases in nixpkgs as well I believe, where we specifically place shared libraries into bin for windows builds. That way (because the bin folders would end up on PATH, it usually just works on windows.
I keep traversing all dependencies, and collect .dlls from lib and bin and place them alongside the final executable, as that seems to be the most reliable way to build some form of distributable for windows.
There was a problem hiding this comment.
So the way I think of it is that the libraries primarily should live in lib, so that you don't pick up extra exe dependencies in the wrong multiple outputs, but each exe needs it's dlls symlinked in its bin directory since there is no RUNPATH or RPATH by which it could find them.
lib is still the cannonical location, but the bin symlinks ensure the exes run.
There was a problem hiding this comment.
Err, why use pre-compiled anything? We are entirely building from source here (except for linux native boostrapping tools).
There was a problem hiding this comment.
I think my thing with the TODO is fine for now.
Ericson2314
force-pushed
the
mingw-rocksdb
branch
from
30a58c3 to
0f04efa
Compare
|
OK this now builds RocksDB, but references a PR commit that I'll probably have to rebase away because facebook/rocksdb#6027 doesn't pass their CI. Should I just vendor the patch then? |
ofborg
bot
added
the
2.status: merge conflict
(cherry picked from commit 0cd8882)
Can just use single commit from facebook/rocksdb#6076 for rocksdb!
This is sightly faster/simpler.
Ericson2314
force-pushed
the
mingw-rocksdb
branch
from
0f04efa to
ae8a905
Compare
ofborg
bot
removed
the
2.status: merge conflict
| THREAD_MSG := ==> building with threading support | ||
| THREAD_CPP := -DZSTD_MULTITHREAD | ||
| -THREAD_LD := -pthread | ||
| +THREAD_LD := |
Motivation for this change
@angerman nerd sniped me. Got all the debs building, now just need to make rocksdb itself work without
localtime_r.Things done
sandboxinnix.confon non-NixOS linux)nix-shell -p nix-review --run "nix-review wip"./result/bin/)nix path-info -Sbefore and after)Notify maintainers
cc @