pixi
pixi copied to clipboard
prefix-dev
Improvement ideas for `pixi search`
Problem description
In https://github.com/mamba-org/mamba/pull/2823 and https://github.com/mamba-org/mamba/pull/2782 there have been some nice improvements for micromamba search. It would be great if pixi could also get some of the features.
`micromamba search pandas=2.1`
❯ micromamba search pandas=2.1
Getting repodata from channels...
conda-forge/osx-arm64 Using cache
conda-forge/noarch Using cache
qc-internal/osx-arm64 Using cache
qc-internal/noarch Using cache
pandas 2.1.1 py310h5924a0a_0 (+ 6 builds)
─────────────────────────────────────────
Name pandas
Version 2.1.1
Build py310h5924a0a_0
Size 11757 kB
License BSD-3-Clause
Subdir osx-arm64
File Name pandas-2.1.1-py310h5924a0a_0.conda
URL https://*****conda.anaconda.org/conda-forge/osx-arm64/pandas-2.1.1-py310h5924a0a_0.conda
MD5 6122cea23eb5a36b9470622a5df282da
SHA256 39d365ab5e6415342295c63ebbf810eca015b94bd7e87135c7ff74975cd2af6e
Dependencies:
- libcxx >=15.0.7
- numpy >=1.22.4,<2.0a0
- python >=3.10,<3.11.0a0
- python >=3.10,<3.11.0a0 *_cpython
- python-dateutil >=2.8.1
- python-tzdata >=2022a
- python_abi 3.10.* *_cp310
- pytz >=2020.1
Other Versions (1):
Version Build
──────────────────────────────────────
2.1.0 py310h5924a0a_0 (+ 2 builds)
`micromamba search pandas=2.1 -v`
❯ micromamba search pandas=2.1 -v
Getting repodata from channels...
pandas 2.1.1 py310h5924a0a_0
────────────────────────────────────────
Name pandas
Version 2.1.1
Build py310h5924a0a_0
Size 11757 kB
License BSD-3-Clause
Subdir osx-arm64
File Name pandas-2.1.1-py310h5924a0a_0.conda
URL https://*****conda.anaconda.org/conda-forge/osx-arm64/pandas-2.1.1-py310h5924a0a_0.conda
MD5 6122cea23eb5a36b9470622a5df282da
SHA256 39d365ab5e6415342295c63ebbf810eca015b94bd7e87135c7ff74975cd2af6e
Dependencies:
- python_abi 3.10.* *_cp310
- python >=3.10,<3.11.0a0
- python >=3.10,<3.11.0a0 *_cpython
- libcxx >=15.0.7
- numpy >=1.22.4,<2.0a0
- python-dateutil >=2.8.1
- pytz >=2020.1
- python-tzdata >=2022a
Other Builds (9):
Version Build
───────────────────────────
2.1.1 py310h5924a0a_1
2.1.1 py311h9e438b8_1
2.1.1 py39h425d09f_0
2.1.1 py39h425d09f_1
2.1.1 py312h58738bf_1
2.1.1 py311h9e438b8_0
2.1.0 py310h5924a0a_0
2.1.0 py311h9e438b8_0
2.1.0 py39h425d09f_0
I am currently missing in pixi search is the ability to show multiple versions of packages as well as the build strings.
I also think that the best match could display more metadata like with the new implementation of micromamba search.
That looks good! I would glady accept PRs! Maybe @Wackyator ? since you build the initial search?
The wildcard search is fixed in #578 but there is still more we could improve.
- Search for specific versions of a package using the match spec e.g.
pixi search python==3.10.0 - The verbosity flag that @pavelzw suggests to get all versions.
- Print a link to prefix.dev to get even more info on the package.
Possibly more, share your ideas!
pixi search --json would be nice too.
❯ micromamba search pytest=6.2.2 --json
{
"query": {
"query": "pytest =6.2.2*",
"type": "search"
},
"result": {
"msg": "",
"pkgs": [
{
"build": "py38h10201cd_0",
"build_number": 0,
"build_string": "py38h10201cd_0",
"channel": "conda-forge",
"constrains": [
"pytest-faulthandler >=2"
],
"depends": [
"packaging",
"toml",
"iniconfig",
"python_abi 3.8.* *_cp38",
"python >=3.8,<3.9.0a0",
"attrs >=19.2.0",
"py >=1.8.2",
"more-itertools >=4.0.0",
"pluggy >=0.12,<1.0.0a1"
],
"fn": "pytest-6.2.2-py38h10201cd_0.tar.bz2",
"license": "MIT",
"md5": "f8a654a6333435956fefaf51045047fe",
"name": "pytest",
"sha256": "8e9323ce0b47273b2f3df9a5fc1fbc6e9f9897515e88307fcde7589bc3d9de22",
"size": 442348,
"subdir": "osx-arm64",
"timestamp": 1611752118,
"track_features": "",
"url": "https://conda.anaconda.org/conda-forge/osx-arm64/pytest-6.2.2-py38h10201cd_0.tar.bz2",
"version": "6.2.2"
},
{
"build": "py39h2804cbe_0",
"build_number": 0,
"build_string": "py39h2804cbe_0",
"channel": "conda-forge",
"constrains": [
"pytest-faulthandler >=2"
],
"depends": [
"packaging",
"toml",
"iniconfig",
"python_abi 3.9.* *_cp39",
"python >=3.9,<3.10.0a0",
"attrs >=19.2.0",
"py >=1.8.2",
"more-itertools >=4.0.0",
"pluggy >=0.12,<1.0.0a1"
],
"fn": "pytest-6.2.2-py39h2804cbe_0.tar.bz2",
"license": "MIT",
"md5": "dec19445bac6ed9e535744e7f1c5269d",
"name": "pytest",
"sha256": "da5c9d835301451b4093a91463124b2a5b5975da9280fa95b21a1c70c7f25beb",
"size": 442665,
"subdir": "osx-arm64",
"timestamp": 1611752232,
"track_features": "",
"url": "https://conda.anaconda.org/conda-forge/osx-arm64/pytest-6.2.2-py39h2804cbe_0.tar.bz2",
"version": "6.2.2"
}
],
"status": "OK"
}
}
Would it be in-line with the vision of pixi search to make it work nearly identically to conda search?
I was inspired to ask based on this question on discord: https://discord.com/channels/1082332781146800168/1266482869954416843/1266482869954416843
I usually reach for conda search when I'm trying to explore repository content, and having conda search return only one result is unhelpful for that. Conda search's output is much more favorable for me.
$ conda search "python>=3.12"
...
python 3.12.0 hab00c5b_0_cpython conda-forge
python 3.12.1 hab00c5b_0_cpython conda-forge
python 3.12.1 hab00c5b_1_cpython conda-forge
python 3.12.2 hab00c5b_0_cpython conda-forge
python 3.12.3 hab00c5b_0_cpython conda-forge
python 3.12.4 h194c7f8_0_cpython conda-forge
And then I can make this detailed by passing -i:
python 3.12.4 h194c7f8_0_cpython
--------------------------------
file name : python-3.12.4-h194c7f8_0_cpython.conda
name : python
version : 3.12.4
build : h194c7f8_0_cpython
build number: 0
size : 30.6 MB
license : Python-2.0
subdir : linux-64
url : http://lxc-nginx-internal.lan/conda/conda-forge/linux-64/python-3.12.4-h194c7f8_0_cpython.conda
md5 : d73490214f536cccb5819e9873048c92
timestamp : 2024-06-17 10:56:11 UTC
constraints :
- python_abi 3.12.* *_cp312
dependencies:
- bzip2 >=1.0.8,<2.0a0
- ld_impl_linux-64 >=2.36.1
- libexpat >=2.6.2,<3.0a0
- libffi >=3.4,<4.0a0
- libgcc-ng >=12
- libnsl >=2.0.1,<2.1.0a0
- libsqlite >=3.46.0,<4.0a0
- libuuid >=2.38.1,<3.0a0
- libxcrypt >=4.4.36
- libzlib >=1.3.1,<2.0a0
- ncurses >=6.5,<7.0a0
- openssl >=3.3.1,<4.0a0
- readline >=8.2,<9.0a0
- tk >=8.6.13,<8.7.0a0
- tzdata
- xz >=5.2.6,<6.0a0
- pip
[ and so on ]
However, this is in contrast to issue #970 which suggests that the pixi search should instead be context-dependent, i.e. have different behavior within pixi projects than without.
So I'd like clarity from maintainers if they have a preference for what this feature shapes out to before considering contributing a solution that may be rejected immediately!
Over in #1827 I posted that the wildcard search seems to be a little too fuzzy, pixi search "gcc*" for example gives a huge number of unrelated results
I would also suggest that pixi search can find substrings.
As of now, it seems limiting that pylsp will not find pylsp-mypy and pylsp-rope for some reason.
I found that conda-forge can do that (as most search engines), and maybe we can connect to its API.
https://conda-forge.org/packages/
Connecting via API is a bad idea as this is not standardized anywhere and it won't work with mirrors like artifactory, prefix.dev, ...
Also, most things are already doable using the rattler_repodata_gateway crate.
Couldn't it be automatically infered, as its in almost every other search engine as well?
I think we should lean into matchspecs (especially once https://github.com/conda/rattler/pull/1480 is implemented) as they are a well defined standard in the conda ecosystem that allows you to specify packages. Once they get glob support in rattler, we should be able to query everything through matchspecs.