Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[V3 Downloader] Update repo without cogs, update single cog #2527

Closed
1 of 3 tasks
jack1142 opened this issue Mar 18, 2019 · 4 comments · Fixed by #2571
Closed
1 of 3 tasks

[V3 Downloader] Update repo without cogs, update single cog #2527

jack1142 opened this issue Mar 18, 2019 · 4 comments · Fixed by #2571
Labels
Type: Feature

Comments

@jack1142
Copy link
Member

@jack1142 jack1142 commented Mar 18, 2019

Feature request

Select the type of feature you are requesting:

  • Cog

  • Command

  • API functionality

Describe your requested feature

I would want to see [p]repo update command which would update repo without updating any cog. This could be useful when new cog is released in repo and you would want to install it without updating other cogs.

From what I've seen to make this change without breaking [p]cog update it would be required to know status of the installed cogs (if they are up-to-date) as currently updating cog is based on checking if the repo is up-to-date and if it isn't downloader updates all cogs from that repo.

Once we know the status of the installed cogs, it probably wouldn't be hard to add functionality to update individual cogs without updating other from the repo, so I thought I add this to the issue as well.

@zephyrkul
Copy link
Contributor

@zephyrkul zephyrkul commented Mar 18, 2019

Both of these features already exist with V3.

  1. Update repos without updating cogs:
    Simply use [p]cog update and when it asks you if you want to reload, react with .

  2. Update only specified cog
    [p]cog update [cog_name]

@jack1142
Copy link
Member Author

@jack1142 jack1142 commented Mar 18, 2019

  1. But in reality cogs are updated and just not reloaded, which means that in next reload or restart of a bot, new version will be used, so it doesn't work as I suggested above.

  2. This doesn't update one cog, it updates every cog from the repo this cog is in.

@mikeshardmind
Copy link
Contributor

@mikeshardmind mikeshardmind commented Mar 18, 2019

I'd love to see this as part of the core behavior, including pinning cogs to specific versions, but it's not here right now. The underlying behavior of this is also why when an author updates a single cog, every installed and loaded cog shows as updated.

@jack1142
Copy link
Member Author

@jack1142 jack1142 commented Mar 18, 2019

Yeah, pinning cogs would go well with these changes too, true. I was wondering, how it should be handled, to make those changes possible. Everything should probably be based on commit ids as by this we could:

  • install cog from specific commit
  • pin cogs to its current state/commit (maybe also option to upgrade and then pin?)
  • upgrade cog to particular commit (which needs to be newer than the one it's currently on)
  • obviously upgrade individual cogs
  • show cogs that can be updated (this would require using git diff, which might not be good performance-wise though, there's also git diff-index, which should faster option, but I'm not git expert, so there might still be some better option)

Commit ids for each cog should be kept in conf's installed variable as we will need to operate on it in some of those functionalities.
I'm interested in helping with this, I'm not sure, if I'll be able to write all of this on my own, but I surely wanted to write my ideas on what could be implemented as it might be useful for someone else too.

And if someone thinks that something from what I said should be handled differently, it would be better to know that too :D

@Twentysix26 Twentysix26 added the Type: Feature label Mar 20, 2019
jack1142 added a commit to jack1142/Red-DiscordBot that referenced this issue Nov 7, 2019
…ommit)

- Repo and Installable have commit property now
- New class inheriting from Installable -
InstalledCog (old one from converters.py removed)
- New Repo.checkout() method, which is also async ctx manager
ref Cog-Creators#2527
jack1142 added a commit to jack1142/Red-DiscordBot that referenced this issue Nov 7, 2019
The most important part of issue Cog-Creators#2527 has been added here
- `[p]repo update` command added
- new conf format - nested dictionary repo_name->cog_name->cog_json
  installed libraries are now kept in conf too
  - `InstalledCog` renamed to `InstalledModule` - installed libraries use this class
  - `Downloader.installed_libraries()` and `Downloader.installed_modules()` added
  - `Downloader._add_to_installed()` and `Downloader._remove_from_installed()`
    now accept list of modules, of both cogs and libraries
- `[p]cog install` tells about fails of copying cog and installing shared libraries
- `[p]cog update` will truly update only chosen cogs (if provided) or cogs that need update
  - pinned cogs aren't checked
  - before update, repos are updated
  - to determine if update is needed `Repo.get_modified_modules()` is used
- `[p]cog pin` and `[p]cog unpin` commands for pinning/unpinning cogs added
- `Repo.checkout()` allows to choose ctx manager exit's checkout revision
- `Repo.install_cog()` returns `InstalledModule` now and raises CopyingError (maybe breaking?)
- `Repo.install_libraries()` returns 2-tuple of installed and failed libraries (maybe breaking?)
- `RepoManager.get_all_cogs()` added, which returns cogs from all repos
- `RepoManager.repos` property added, which contains tuple of `Repo`
mikeshardmind pushed a commit that referenced this issue Nov 8, 2019
* feat(downloader): Install cog from specific commit in repo (initial commit)

- Repo and Installable have commit property now
- New class inheriting from Installable -
InstalledCog (old one from converters.py removed)
- New Repo.checkout() method, which is also async ctx manager
ref #2527

* fix(downloader): Keep information about repo's branch in config

- This is needed to make sure that repo can go back from detached state in some rare unexpected
cases
- current branch is determined by `git symbolic-ref` now as this command errors for detached
HEAD

* feat(downloader): Update repo without cogs, update single cog

The most important part of issue #2527 has been added here
- `[p]repo update` command added
- new conf format - nested dictionary repo_name->cog_name->cog_json
  installed libraries are now kept in conf too
  - `InstalledCog` renamed to `InstalledModule` - installed libraries use this class
  - `Downloader.installed_libraries()` and `Downloader.installed_modules()` added
  - `Downloader._add_to_installed()` and `Downloader._remove_from_installed()`
    now accept list of modules, of both cogs and libraries
- `[p]cog install` tells about fails of copying cog and installing shared libraries
- `[p]cog update` will truly update only chosen cogs (if provided) or cogs that need update
  - pinned cogs aren't checked
  - before update, repos are updated
  - to determine if update is needed `Repo.get_modified_modules()` is used
- `[p]cog pin` and `[p]cog unpin` commands for pinning/unpinning cogs added
- `Repo.checkout()` allows to choose ctx manager exit's checkout revision
- `Repo.install_cog()` returns `InstalledModule` now and raises CopyingError (maybe breaking?)
- `Repo.install_libraries()` returns 2-tuple of installed and failed libraries (maybe breaking?)
- `RepoManager.get_all_cogs()` added, which returns cogs from all repos
- `RepoManager.repos` property added, which contains tuple of `Repo`

* test(downloader): Repo.current_branch() throws an exception, when branch can't be determined

* style(downloader): rename _add_to_installed to _save_to_installed

This method is used for both adding and updating existing modules in Config

* refactor(downloader): add ctx.typing() for few commands

`[p]cog install` is nested hell, can't wait for moving install logic to separate method

* fix(downloader): refactor and fix `set` usage

* perf(downloader): update commits for ALL checked modules to omit diffs next time

This will also disable running git diff for cogs that have the same commit as the latest one

* style(downloader): few style improvements

- use of mutable object in method definition
- make Repo._get_full_sha1() public method
- too long
line
- don't use len to check if sequence is empty

* feat(downloader): add `[p]cog updateallfromrepos` and `[p]cog updatetoversion` commands

- moved cog update logic into `Downloader._cog_update_logic()` (lack of better name)
  - splitted
whole cog update process into smaller methods
  - might still need some improvements
- added new
methods to `Repo` class:
  - `is_on_branch()` to check if repo is currently checked out to branch

- `is_ancestor()` to check if one commit is ancestor of the other
- fix for
`Downloader._available_updates()` behaviour
broken by commit
5755ab08ba67556b3863e907c6f44d80f4f13d88

* feat(downloader): try to find last commit where module is still present

Enhancements:
- `Installable` now has `repo` attribute containing repo object or `None` if repo is
missing
- `Downloader._install_cogs()` and `Downloader._reinstall_libraries()` are able to install
modules from different commits of repo
- `Repo.checkout()` as ctx manager will now exit to commit
which was active before checking out
- unification of `rev` and `hash` terms:
All function
parameters are explicitly called `hash`, if it can only be commit's full sha1 hash or `rev` if it
can be anything that names a commit object, see
[link](https://git-scm.com/docs/git-rev-parse#_specifying_revisions)
- new
`Repo.get_last_module_occurence()` method, which gets module's Installable from last commit in which
it still occurs

* docs(downloader): Add basic description for `InstalledModule`

* fix(downloader): cog ignored during updates if its commit was missing

After config format update, commit string is empty until update and when such cog was checked and it
wasn't available in repo anymore, it was ignored

* refactor(downloader): Installing cogs from specific rev will pin them

* perf(downloader): Don't checkout when current commit equals target hash

- changes to `Repo.checkout()`:
  - `exit_to_rev` is now keyword only argument
  - added
`force_checkout` to force checkout even if `Repo.commit` value is the same as target hash

* refactor(downloader): Repo._run() stderr is redirected to debug log now

- added two keyword arguments:
  - `valid_exit_codes` which specifies valid exit codes, used to
determine if stderr should be sent as debug or error level in logging
  - `debug_only` which
specifies if stderr can be sent only as debug level in logging

* style(downloader): stop using `set` as arg name in `_load_repos()`

* feat(downloader): pass multiple cogs to `[p]cog (un)pin`

* refactor(downloader): accept module name instead of instance, fix spelling

* style(downloader): few small style changes

* fix(downloader): add type annotations + fixes based on them

- fix wrong type annotations and add a lot of new ones
- add checks for `Installable.repo` being `None`
- fix wrong return type in `Downloader._install_requirements`
- show repo names correctly when updating all repos
- fix error when some requirement fails to install

BREAKING CHANGE:
- type of `Repo.available_modules` is now consistent (always `tuple`)

* tests: use same event loop policy as in Red's code

* enhance(downloader): fully handle ambiguous revisions

* build(deps): add pytest-mock dependency to tests extra

* fix(downloader): minor fixes

* feat(downloader): add tool for editing Downloader's test repo

This script aims to help update the human-readable version of repo
used for git integration tests in ``redbot/tests/downloader_testrepo.export``
by exporting/importing it in/from provided directory.

Note
----
Editing `downloader_git_test_repo.export` file manually is strongly discouraged,
especially editing any part of commit directives as that causes a change in the commit's hash.
Another problem devs could encounter when trying to manually edit that file
are editors that will use CRLF instead of LF for new line character(s) and therefore break it.

I also used `.gitattributes` to prevent autocrlf from breaking testrepo.

Also, if Git ever changes currently used SHA-1 to SHA-256 we will have to
update old hashes with new ones. But it's a small drawback,
when we can have human-readable version of repo.

Known limitations
-----------------
``git fast-export`` exports commits without GPG signs so this script disables it in repo's config.
This also means devs shouldn't use ``--gpg-sign`` flag in ``git commit`` within the test repo.

* tests(downloader): add git tests and test repo for them

Also added Markdown file that is even more clear than export file
on what the test repo contains.
This is manually created but can be automated on later date.

* test(downloader): add more tests related to RepoManager

These tests use expected output that is already guaranteed by git tests.

* chore(CODEOWNERS): add jack1142 to Downloader's folders

I know this doesn't actually give any benefit to people that don't have
write permission to the repo but I saw other big fella devs doing this,
so I think this might be advisable.

* enhance(downloader): allow easy schema updates in future

* enhance(downloader): more typing fixes, add comments for clarity

* feat(downloader): add python and bot version check to update process

follow-up on #2605, this commit fully fixes #1866

* chore(changelog): add towncrier entries

* fix(downloader): use `*args` instead of `commands.Greedy`

* fix(downloader): hot-reload issue - `InstallableType` now inherits from `IntEnum`

There's desync of `InstallableType` class types due to hot-reload
and `IntEnum` allows for equality check between different types

* enhance(downloader): ensure there's no cog with same name installed

should fix #2927

* fix(downloader): last few changes before marking as ready for review
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Type: Feature
Projects
None yet
Development

Successfully merging a pull request may close this issue.

4 participants