Closed andrewtavis closed 1 month ago
CC @wkyoshida, in case you have some ideas on this :)
The workflow is failing because one of the dependancies language-data
depends on marisa-trie<0.8.0 and >=0.7.7
.
And marisa-trie<0.8.0 and >=0.7.7
does not have support for Python 3.12
which is the error in the CI logs
.
As mentioned in the marisa-trie docs, marisa-trie 1.1.0
has added support for python 3.12 .
The solution is to remove the language-data
dependancy from requirements.txt
as I think it is not being used anywhere in the project.
I also tried pinning both language-data
and marisa-trie
to their latest versions but this won't work beacuse of this:
ERROR: Cannot install language-data==1.1 and marisa-trie>=1.1.0 because these package versions have conflicting dependencies.
The user requested marisa-trie>=1.1.0
language-data 1.1 depends on marisa-trie<0.8.0 and >=0.7.7
To fix this you could try to:
1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflict
ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/topics/dependency-resolution/#dealing-with-dependency-conflicts```
@andrewtavis @wkyoshida
Thanks for looking into this, @shashank-iitbhu! 😊 I just checked and the following is in the pyproject.toml for langcodes:
[tool.poetry.dependencies]
python = ">= 3.6"
language-data = { version = "^1.1", optional = true }
[tool.poetry.dev-dependencies]
language-data = { version = "^1.1", optional = true }
[tool.poetry.extras]
data = ["language-data"]
Do you want to give try to remove language-data
from the dependencies and hopefully it'll work, @shashank-iitbhu? We can also just not support Python 3.12 right now. It seems like there was a recent fork of language-data that's now maintained :)
It looks like the new language-data
will be just the lower bound of marisa-trie
, but then it seems to be already in production. Let's give unpinning a try and focus on testing Python 3.9 as we are now.
Thinking about this a bit more and what a hassle langcodes
/language-data
and marisa-trie
have been since being added in, would it make sense to revert this and shift back to a dictionary that we're maintaining based on the languages we have the in extract_transform/languages
directory? We just wanted this for language to ISO-2 conversion.
CC @wkyoshida
@shashank-iitbhu, would you want to implement a reverse of #55 such that we have the codes saved for all the current languages in the languages
directory? We'd be adding those in to language_meta_data.json :)
Sure @andrewtavis, will open a PR for this.
Thanks, @shashank-iitbhu!
Closed by #129. Thanks for this, @shashank-iitbhu!
Reopening here and removing the mac build for now as this is once again problematic...
Well would you look at them Mac builds! The combination of the following got this fixed:
I referenced the following Stack Overflow answer when I was setting up the project again from scratch and realized that it likely could help with the problems we were having here 😊
Closing until the next time the Mac builds break! 😅
Terms
Behavior
As of now the build for all versions of MacOS is failing. This was originally because of Tensorflow, which has now been removed as a dependency in favor of other packages that were also being used and had some similar functionality (parallel downloads). The most recent error is:
Would be great if we could figure this out!