| 10 Jun 2024 |
|  NixOS Moderation Bot unbanned  @jonringer:matrix.org. | 00:17:07 |
 willbush | I also missed :(. I have had other things come up, but I still want to help out. I'm hoping to be more active next month. Feel free to reschedule Jit.si meet without my crab fit schedule :) | 22:03:26 |
| 11 Jun 2024 |
 Alexandros Liarokapis | Hi all. Not sure if this is the proper Room to discuss this. I was thinking, maybe it would be worth enabling patchelf to change hard-coded paths in the text segment to relative ones, then provide symlinks in the $out folder.
This is an alternative to builfFHSEnv that should also work on Darwin (not with patchelf but similar)
not all of them obviously, something like by providing a map of transforms
it would also help with manual dlopens | 10:01:42 |
 Minijackson | Can it work? Doesn't this mess up pointers to the text segment? | 11:10:54 |
 nbp | Most of the time no. | 12:04:01 |
| nbp | * (patching path of different length) Most of the time no. | 12:04:30 |
| Minijackson | how so? I would assume that the program knows the start address and the length of the string, so we can't put longer strings without having to shift the rest, and shorter strings aren't guaranteed to work either | 12:08:12 |
| nbp | the program yes, the patching tool probably does not. | 12:23:58 |
| nbp | then you have different encoding, depending on the language, and other variables that would have to be shifted otherwise, assumption on the addresses … | 12:25:46 |
| Minijackson | In reply to @nbp:mozilla.org
the program yes, the patching tool probably does not. I'm not sure I understand this sentence. Isn't the point patching programs? | 12:32:24 |
| nbp | Well patchelf is patching the header of the ELF binary, but it is assumed unsafe to attempt patching random strings, unless you are sure that they have the same length. | 12:37:40 |
| nbp | The replaceDependencies is the closest thing there is to replacing raw path to the nix store by another one in the nix-store, but beware that if the version changes from 0.4.0 to 0.4.0a this would cause a lot of issues. | 12:39:24 |
| Minijackson | so for the solution proposed by Alexandros Liarokapis, unless the paths are of the same length, there's quite a chance that it would lead to bug, then | 12:44:24 |
| Alexandros Liarokapis | Converting longer paths to shorter relative ones should work | 12:44:27 |
| Alexandros Liarokapis | Hmm yea I guess it depends on the encoding | 12:44:46 |
| Alexandros Liarokapis | Probably only works for c strings | 12:44:52 |
| Minijackson | if the path are shorter, it could be padded with /, but seems hacky x) | 12:44:57 |
| Minijackson | In reply to @aliarokapis:matrix.org
Probably only works for c strings even then, depending on compiler optimisations and other factors, the length of the string might be assumed | 12:45:33 |
| Alexandros Liarokapis | Padding with null strings is also possible | 12:45:34 |
| Alexandros Liarokapis | * Padding with null chars is also possible | 12:45:49 |
| Minijackson | In reply to @aliarokapis:matrix.org
Padding with null chars is also possible not sure, if the compiler optimises printf("some string") to write(stdout, "some string", 11), replacing some string with something shorter but null padded will print null bytes to stdout | 12:46:43 |
| nbp | In reply to @aliarokapis:matrix.org
Converting longer paths to shorter relative ones should work No, Pascal for example encode the length next to the string. | 12:53:00 |
| nbp | Some compilers could bake-in strlen of constants at compile time. | 12:55:12 |
| willbush | infinisil: Philip Taron (UTC-8): I've been kinda wanting a place to share and discuss Rust blog posts / news with you guys. Any ideas? | 13:08:56 |
 @aloisw:kde.org | In reply to @nbp:mozilla.org
The replaceDependencies is the closest thing there is to replacing raw path to the nix store by another one in the nix-store, but beware that if the version changes from 0.4.0 to 0.4.0a this would cause a lot of issues. This is why it enforces the same length already. | 14:10:04 |
 Philip Taron (UTC-8) | I'm in most of the Matrix rooms, and on NixOS discourse. If you want something else, my email is philip.taron@gmail.com 🙂 | 14:57:17 |
| 14 Jun 2024 |
 infinisil | Weekly office hour now :) https://meet.jit.si/nixpkgs-architecture | 17:57:50 |
| infinisil | Cc willbush Philip Taron (UTC-8) Dominic Mills | 17:59:13 |
| infinisil | With Will and Dominic in the call I created a bunch of PRs improve some misc stuff in Nixpkgs, mostly churning this issue down :) | 19:50:53 |
| infinisil | If anybody wants to help out, feel free to fix more cases mentioned in that issue :D | 19:51:31 |
