| 12 Jan 2026 |
 Ilan Joselevich (Kranzes) | 
Download clipboard.png | 18:34:37 |
| Ilan Joselevich (Kranzes) | It feels so wrong having to reimplement the broken behavior of NIX_PATH parsing | 18:34:39 |
| Ilan Joselevich (Kranzes) | If it was "correct" it should've had all the 5 variants, not just 4, remote flakes are broken in NIX_PATH, and so are URLs with colons (not sure if they even work in -I) | 18:36:40 |
 Robert Hensing (roberth) | I'm glad you've joined us in our suffering | 18:36:48 |
| Robert Hensing (roberth) | welcome to the club | 18:36:52 |
| Ilan Joselevich (Kranzes) | Also the C++ implementation is very very misleading, it has this "isPesudoUrl" function, that's not even being used correctly, and suggets there are far more schema types like "ssh" or "s3" | 18:37:56 |
 raitobezarius | but you should probably follow this idea and implement it correctly | 18:38:16 |
| Ilan Joselevich (Kranzes) | this function came originally form libstore 9 years ago, but was moved because it's only used for NIX_PATH, but that doesnt even make sense, because it has things that dont even work on NIX_PATH | 18:38:40 |
| Ilan Joselevich (Kranzes) | for example, the ssh thing, if you put an ssh URI, i get an error from curl about unsupported protocol | 18:39:12 |
| raitobezarius | and if you put a finger url… | 18:39:41 |
| Ilan Joselevich (Kranzes) | there's also funny errors from S3 too | 18:40:00 |
 Sergei Zimmerman (xokdvium) | What also worries be a bit in terms of snix compat is snix is using the url crate which is the whatwg one and libcurl nowadays implements rfc one | 18:40:03 |
| Ilan Joselevich (Kranzes) | depending on Nix versions and Lix versions | 18:40:07 |
| Sergei Zimmerman (xokdvium) | But actually we are not using the strict curl API so anything goes... | 18:40:48 |
| Sergei Zimmerman (xokdvium) | But flakes are kinda 3986 but not really... | 18:41:06 |
| Ilan Joselevich (Kranzes) | tfw implementing correctly = backwards incompatible with nix | 18:41:32 |
| raitobezarius | no | 18:41:39 |
| raitobezarius | schemes should be generalized and pluggable | 18:41:51 |
| Robert Hensing (roberth) | this part is still experimental fwiw | 18:42:12 |
| Robert Hensing (roberth) | (flakerefs) | 18:42:58 |
| Robert Hensing (roberth) | or is it, because Eelco poked a NIX_PATH hole? | 18:43:18 |
| Robert Hensing (roberth) | :( | 18:43:22 |
| raitobezarius | well, if you put a flakeref in a NIX_PATH, it doesn't require an xp feature | 18:43:31 |
| raitobezarius | but at the boundary, an xp feature will be required | 18:43:35 |
| Robert Hensing (roberth) | it damn well should have... | 18:43:47 |
| raitobezarius | it doesn't matter much in the end | 18:44:06 |
| raitobezarius | flakerefs are theoretically experimental | 18:44:18 |
| raitobezarius | in practice, breaking backward compat of it is not a good idea | 18:44:31 |
| Robert Hensing (roberth) | they're already part of the sediment record | 18:44:40 |
| raitobezarius | flakerefs as a term should die | 18:44:46 |
