Don't depend on flake8 at runtime #6830

deivid-rodriguez · 2023-03-14T11:51:55Z

It should avoid avoidable resolution issues when installing old Python's
dynamically.

jeffwidman

I hadn't realized we were installing this in our prod requirements.

python/script/ci-test

python/helpers/dev-requirements.txt

python/script/ci-test

Instead move running flake8 to our lint job and keep that dependency on a `requirements.txt` file in the repository root.

I realized I forgot to update the `.python-version` file to `3.8`, so we're still installing an older `flake8` version. And it's also breaking when

runs on `dependabot-core` because it sees this `3.7` and says "that's not a supported version!" Furthermore, we haven't configured

to watch the `./requirements.txt` which is separate from `./python/helpers/requirements.txt`. I started to update both of those but the complexity here didn't smell right. This way of installing `flake8` is actually different than linters for other native helpers, where we just include them in `composer.json`, `package.json`, etc. So the intuitive thing is to install `flake8` into `./python/helpers/requirements.txt`. In fact we used to do this, but in #6830 / f407685 we switched to this new method because `flake8` dropped `3.6` support before we were ready to do that here in

. Given that in

we're moving toward aligning our support policies with upstream Python... ie, when they EOL a version, we will also drop support for it, then the risk of having a `flake8` version that's dropping support for a Python version that isn't yet EOL is quite small. Interestingly, `flake8` was very slow to drop `3.6`, but when they did, the maintainer also dropped `3.7` to save himself time even though it was a few months before it was EOL'd. So we could hit this scenario again... but I think the odds of that are somewhat low given the amount of blowback the maintainer got about dropping 3.7 before it was EOL'd. And regardless the consequences are also low... we'd simply revert that `flake8` version bump and temporarily hold it back for a few months until the upstream python version was dropped. Overall this feels like six one way, half a dozen the other as either way we are protecting against something. But switching back to how we do it for our other native helpers is both much simpler setup and also more internally consistent. So let's solve this by reverting back to the simpler way of installing the linter alongside the helper. As an aside, restoring this simplicity is yet another win from us choosign to align with upstream Python's EOL policy. This reverts commit f407685.

I realized I forgot to update the `.python-version` file to `3.8`, so we're still installing an older `flake8` version. And it's also breaking when

runs on `dependabot-core` because it sees this `3.7` and says "that's not a supported version!" Furthermore, we haven't configured

to watch the `./requirements.txt` which is separate from `./python/helpers/requirements.txt`. I started to update both of those but the complexity here didn't smell right. This way of installing `flake8` is actually different than linters for other native helpers, where we just include them in `composer.json`, `package.json`, etc. So the intuitive thing is to install `flake8` into `./python/helpers/requirements.txt`. In fact we used to do this, but in #6830 / f407685 we switched to this new method because `flake8` dropped `3.6` support before we were ready to do that here in

. Given that in

we're moving toward aligning our support policies with upstream Python... ie, when they EOL a version, we will also drop support for it, then the risk of having a `flake8` version that's dropping support for a Python version that isn't yet EOL is quite small. Interestingly, `flake8` was very slow to drop `3.6`, but when they did, the maintainer also dropped `3.7` to save himself time even though it was a few months before it was EOL'd. So we could hit this scenario again... but I think the odds of that are somewhat low given the amount of blowback the maintainer got about dropping 3.7 before it was EOL'd. And regardless the consequences are also low... we'd simply revert that `flake8` version bump and temporarily hold it back for a few months until the upstream python version was dropped. Overall this feels like six one way, half a dozen the other as either way we are protecting against something. But switching back to how we do it for our other native helpers is both much simpler setup and also more internally consistent. So let's solve this by reverting back to the simpler way of installing the linter alongside the helper. As an aside, restoring this simplicity is yet another win from us choosing to align with upstream Python's EOL policy. This reverts commit f407685.

I realized I forgot to update the `.python-version` file to `3.8`, so we're still installing an older `flake8` version. And it's also breaking when

runs on `dependabot-core` because it sees this `3.7` and says "that's not a supported version!" Furthermore, we haven't configured

to watch the `./requirements.txt` which is separate from `./python/helpers/requirements.txt`. I started to update both of those but the complexity here didn't smell right. This way of installing `flake8` is actually different than linters for other native helpers, where we just include them in `composer.json`, `package.json`, etc. So the intuitive thing is to install `flake8` into `./python/helpers/requirements.txt`. In fact we used to do this, but in #6830 / f407685 we switched to this new method because `flake8` dropped `3.6` support before we were ready to do that here in

. Given that in

we're moving toward aligning our support policies with upstream Python... ie, when they EOL a version, we will also drop support for it, then the risk of having a `flake8` version that's dropping support for a Python version that isn't yet EOL is quite small. Interestingly, `flake8` was very slow to drop `3.6`, but when they did, the maintainer also dropped `3.7` to save himself time even though it was a few months before it was EOL'd. So we could hit this scenario again... but I think the odds of that are somewhat low given the amount of blowback the maintainer got about dropping 3.7 before it was EOL'd. And regardless the consequences are also low... we'd simply revert that `flake8` version bump and temporarily hold it back for a few months until the upstream python version was dropped. Overall this feels like six one way, half a dozen the other as either way we are protecting against something. But switching back to how we do it for our other native helpers is both much simpler setup and also more internally consistent. So let's solve this by reverting back to the simpler way of installing the linter alongside the helper. As an aside, restoring this simplicity is yet another win from us choosing to align with upstream Python's EOL policy. This reverts commit f407685.

I realized I forgot to update the `.python-version` file to `3.8`, so we're still installing an older `flake8` version. And it's also breaking when

runs on `dependabot-core` because it sees this `3.7` and says "that's not a supported version!" Furthermore, we haven't configured

to watch the `./requirements.txt` which is separate from `./python/helpers/requirements.txt`. I started to update both of those but the complexity here didn't smell right. This way of installing `flake8` is actually different than linters for other native helpers, where we just include them in `composer.json`, `package.json`, etc. So the intuitive thing is to install `flake8` into `./python/helpers/requirements.txt`. In fact we used to do this, but in #6830 / f407685 we switched to this new method because `flake8` dropped `3.6` support before we were ready to do that here in

. Given that in

we're moving toward aligning our support policies with upstream Python... ie, when they EOL a version, we will also drop support for it, then the risk of having a `flake8` version that's dropping support for a Python version that isn't yet EOL is quite small. Interestingly, `flake8` was very slow to drop `3.6`, but when they did, the maintainer also dropped `3.7` to save himself time even though it was a few months before it was EOL'd. So we could hit this scenario again... but I think the odds of that are somewhat low given the amount of blowback the maintainer got about dropping 3.7 before it was EOL'd. And regardless the consequences are also low... we'd simply revert that `flake8` version bump and temporarily hold it back for a few months until the upstream python version was dropped. Overall this feels like six one way, half a dozen the other as either way we are protecting against something. But switching back to how we do it for our other native helpers is both much simpler setup and also more internally consistent. So let's solve this by reverting back to the simpler way of installing the linter alongside the helper. As an aside, restoring this simplicity is yet another win from us choosing to align with upstream Python's EOL policy. This reverts commit f407685.

I realized I forgot to update the `.python-version` file to `3.8`, so we're still installing an older `flake8` version. And it's also breaking when

runs on `dependabot-core` because it sees this `3.7` and says "that's not a supported version!" Furthermore, we haven't configured

to watch the `./requirements.txt` which is separate from `./python/helpers/requirements.txt`. I started to update both of those but the complexity here didn't smell right. This way of installing `flake8` is actually different than linters for other native helpers, where we just include them in `composer.json`, `package.json`, etc. So the intuitive thing is to install `flake8` into `./python/helpers/requirements.txt`. In fact we used to do this, but in #6830 / f407685 we switched to this new method because `flake8` dropped `3.6` support before we were ready to do that here in

. Given that in

we're moving toward aligning our support policies with upstream Python... ie, when they EOL a version, we will also drop support for it, then the risk of having a `flake8` version that's dropping support for a Python version that isn't yet EOL is quite small. Interestingly, `flake8` was very slow to drop `3.6`, but when they did, the maintainer also dropped `3.7` to save himself time even though it was a few months before it was EOL'd. So we could hit this scenario again... but I think the odds of that are somewhat low given the amount of blowback the maintainer got about dropping 3.7 before it was EOL'd. And regardless the consequences are also low... we'd simply revert that `flake8` version bump and temporarily hold it back for a few months until the upstream python version was dropped. Overall this feels like six one way, half a dozen the other as either way we are protecting against something. But switching back to how we do it for our other native helpers is both much simpler setup and also more internally consistent. So let's solve this by reverting back to the simpler way of installing the linter alongside the helper. As an aside, restoring this simplicity is yet another win from us choosing to align with upstream Python's EOL policy. This reverts commit f407685.

…bot#7836) I realized I forgot to update the `.python-version` file to `3.8`, so we're still installing an older `flake8` version. And it's also breaking when

runs on `dependabot-core` because it sees this `3.7` and says "that's not a supported version!" Furthermore, we haven't configured

to watch the `./requirements.txt` which is separate from `./python/helpers/requirements.txt`. I started to update both of those but the complexity here didn't smell right. This way of installing `flake8` is actually different than linters for other native helpers, where we just include them in `composer.json`, `package.json`, etc. So the intuitive thing is to install `flake8` into `./python/helpers/requirements.txt`. In fact we used to do this, but in dependabot#6830 / f407685 we switched to this new method because `flake8` dropped `3.6` support before we were ready to do that here in

. Given that in

we're moving toward aligning our support policies with upstream Python... ie, when they EOL a version, we will also drop support for it, then the risk of having a `flake8` version that's dropping support for a Python version that isn't yet EOL is quite small. Interestingly, `flake8` was very slow to drop `3.6`, but when they did, the maintainer also dropped `3.7` to save himself time even though it was a few months before it was EOL'd. So we could hit this scenario again... but I think the odds of that are somewhat low given the amount of blowback the maintainer got about dropping 3.7 before it was EOL'd. And regardless the consequences are also low... we'd simply revert that `flake8` version bump and temporarily hold it back for a few months until the upstream python version was dropped. Overall this feels like six one way, half a dozen the other as either way we are protecting against something. But switching back to how we do it for our other native helpers is both much simpler setup and also more internally consistent. So let's solve this by reverting back to the simpler way of installing the linter alongside the helper. As an aside, restoring this simplicity is yet another win from us choosing to align with upstream Python's EOL policy. This reverts commit f407685.

deivid-rodriguez force-pushed the deivid-rodriguez/remove-runtime-dep-on-flake8 branch from a90cfc2 to 07b1da9 Compare March 14, 2023 12:34

jeffwidman reviewed Mar 14, 2023

View reviewed changes

python/script/ci-test Show resolved Hide resolved

python/helpers/dev-requirements.txt Outdated Show resolved Hide resolved

python/script/ci-test Outdated Show resolved Hide resolved

deivid-rodriguez force-pushed the deivid-rodriguez/remove-runtime-dep-on-flake8 branch 5 times, most recently from 099f91d to c2aa020 Compare March 14, 2023 17:19

deivid-rodriguez marked this pull request as ready for review March 14, 2023 17:19

deivid-rodriguez requested a review from a team as a code owner March 14, 2023 17:19

mctofu approved these changes Mar 14, 2023

View reviewed changes

deivid-rodriguez force-pushed the deivid-rodriguez/remove-runtime-dep-on-flake8 branch from c2aa020 to 3534e76 Compare March 22, 2023 17:56

mctofu approved these changes Mar 23, 2023

View reviewed changes

deivid-rodriguez force-pushed the deivid-rodriguez/remove-runtime-dep-on-flake8 branch 2 times, most recently from f8dcdf4 to af7d9fa Compare March 24, 2023 20:00

jeffwidman mentioned this pull request Mar 24, 2023

Re-evaluate whether we should bundle linters for native helpers into the code #6908

Open

1 task

deivid-rodriguez force-pushed the deivid-rodriguez/remove-runtime-dep-on-flake8 branch 4 times, most recently from c848d30 to 0368125 Compare March 27, 2023 18:18

Don't depend on flake8 at runtime

f112417

Instead move running flake8 to our lint job and keep that dependency on a `requirements.txt` file in the repository root.

deivid-rodriguez force-pushed the deivid-rodriguez/remove-runtime-dep-on-flake8 branch from 0368125 to f112417 Compare March 28, 2023 08:53

deivid-rodriguez merged commit f407685 into main Mar 28, 2023

deivid-rodriguez deleted the deivid-rodriguez/remove-runtime-dep-on-flake8 branch March 28, 2023 10:27

jeffwidman mentioned this pull request Aug 17, 2023

Revert "Don't depend on flake8 at runtime (#6830)" #7836

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Don't depend on flake8 at runtime #6830

Don't depend on flake8 at runtime #6830

deivid-rodriguez commented Mar 14, 2023 •

edited by jeffwidman

jeffwidman left a comment

Don't depend on flake8 at runtime #6830

Don't depend on flake8 at runtime #6830

Conversation

deivid-rodriguez commented Mar 14, 2023 • edited by jeffwidman

jeffwidman left a comment

Choose a reason for hiding this comment

deivid-rodriguez commented Mar 14, 2023 •

edited by jeffwidman