| 22 Jun 2021 |
 pamplemousse | And some things are pretty mysterious...
For example, when using --prefix=outputs/out:
[pamplemousse@wasu:~/Workspace/tools/nix]$ ldd outputs/out/bin/nix-build
[...]
libnixexpr.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixexpr.so (0x00007f1b1fe51000)
libnixmain.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixmain.so (0x00007f1b1fafe000)
libnixfetchers.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixfetchers.so (0x00007f1b1f735000)
libnixstore.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixstore.so (0x00007f1b1ea89000)
libnixutil.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixutil.so (0x00007f1b1e49b000)
libnixcmd.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixcmd.so (0x00007f1b1e2b9000)
[...]
I don't understand how the outputs/out/lib/*.so are referenced here... (i.e. where in the .mk these are set up).
| 23:58:22 |
| 23 Jun 2021 |
 abathur | In reply to @pamplemouss_:matrix.org
And some things are pretty mysterious...
For example, when using --prefix=outputs/out:
[pamplemousse@wasu:~/Workspace/tools/nix]$ ldd outputs/out/bin/nix-build
[...]
libnixexpr.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixexpr.so (0x00007f1b1fe51000)
libnixmain.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixmain.so (0x00007f1b1fafe000)
libnixfetchers.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixfetchers.so (0x00007f1b1f735000)
libnixstore.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixstore.so (0x00007f1b1ea89000)
libnixutil.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixutil.so (0x00007f1b1e49b000)
libnixcmd.so => /home/pamplemousse/Workspace/tools/nix/outputs/out/lib/libnixcmd.so (0x00007f1b1e2b9000)
[...]
I don't understand how the outputs/out/lib/*.so are referenced here... (i.e. where in the .mk these are set up).
not familiar with building nix, but it looks it's fairly abstracted :)
the individual local.mk files scattered through the source get iteratively processed at https://github.com/NixOS/nix/blob/master/mk/lib.mk#L96 and then instantiated a few lines later | 03:13:13 |
| abathur | not certain if that answers the question | 03:13:28 |
| pamplemousse | Yeah, I more or less figured that.
What I am confused about is that as per mk/libraries.mk, it seems that the libraries libnix*.so are built in their respective src/lib*/, and only moved to outputs/out/ (or whatever is in --prefix) when make install (mk/lib.mk using "functions" declared in mk/install.mk).
I don't get how the final binary is linked agains't it before they are installed π | 03:19:19 |
 Regnat | pamplemousse as far as I understand, the link logic is actually duplicated: make will build everything in place (and src/nix/nix will use the libs built in-tree), while make install will directly link things at their final location, and set the rpath to this location (here) | 05:17:47 |
| Regnat | So running make and make install will actually result in two different builds of the .so files | 05:19:09 |
|  anubhavkini joined the room. | 06:52:53 |
 Jan Tojnar | Any chance of getting https://github.com/NixOS/nix/pull/4873 in? | 15:38:50 |
| pamplemousse | I didn't think about the rpathπ€¦ββοΈ | 16:01:53 |
| pamplemousse | In reply to @theophane:hufschmitt.net
So running make and make install will actually result in two different builds of the .so files Do you mean that the .so will be compiled twice? | 16:03:00 |
| Regnat | In reply to @pamplemouss_:matrix.org
Do you mean that the .so will be compiled twice? I think so, yes | 16:53:59 |
| Regnat | (I mean, if you just run make install then the in-tree so wonβt be compiled, but if you run both, then they will be compiled twice) | 16:54:33 |
| pamplemousse | π€ ...
I thought they were only compiled on make, and mved on make install.
Rest is rpath doing its job at run-time, and being set to the proper destination on make install... | 16:55:24 |
| abathur | I guess --trace would disambiguate, if it's digestible | 18:39:02 |
| pamplemousse | --just-print | 18:41:49 |
| pamplemousse | does not even run the commands :) | 18:42:11 |
|  Ambroisie joined the room. | 19:13:32 |
| 24 Jun 2021 |
 niksnut | There is basically no reason anymore to run make, just use make install. This is because the tests require an installed package. | 11:02:22 |
| pamplemousse | Should this be reflected in https://hydra.nixos.org/build/144378897/download/2/manual/contributing/hacking.html ? | 15:31:28 |
| pamplemousse | And also:
no reason anymore
I wonder if my use case is not one such reason...
I want to produce binaries, that use customized (instrumented) version of the different libraries.
| 15:40:26 |
| pamplemousse | * And also:
no reason anymore
I wonder if my use case is not one such reason...
I want to produce binaries, that use customized (instrumented) version of the different libraries.
| 15:40:34 |
| Regnat | In reply to @pamplemouss_:matrix.org
And also:
no reason anymore
I wonder if my use case is not one such reason...
I want to produce binaries, that use customized (instrumented) version of the different libraries.
Iβd rather make a fully different build with different configure flags for that. That would make things more robust | 18:25:45 |
| 25 Jun 2021 |
|  Bramfatur joined the room. | 13:59:29 |
 ma27 | niksnut: is there anything else tbd for https://github.com/NixOS/nix/pull/4770 ? %) | 21:32:42 |
|  -(πeloΟ)- changed their display name from -(GNU/βΟ)- to -(NIX/βΟ)-. | 22:06:01 |
| 26 Jun 2021 |
|  grahamc invited  mjolnir. | 01:26:17 |
| mjolnir joined the room. | 01:26:17 |
| grahamcchanged room power levels. | 01:26:27 |
| 27 Jun 2021 |
|  tomberek joined the room. | 05:02:55 |
| 30 Jun 2021 |
|  balsoft joined the room. | 10:54:08 |
