Hooks

[aljazerzen]

• make hooks a single string, instead of an array,
• invoke /bin/sh instead of the process directly,
• set cwd,
• WSL
• tests

---

• project.init.after
• branch.switch.after
• branch.wipe.after
• migration.apply.after

---

Do we want before hooks too? I've implemented some because they were easy.

• project.init.before
• branch.switch.before
• branch.wipe.before
• migration.apply.before

[aljazerzen]

I need an install of windows to implement hooks in WSL.

[msullivan]

Looks good!

[vpetrovykh]

• project.init.after hook works and triggers migration hooks if there are migrations
• branch.switch.before and branch.switch.after hooks respond to switch, but they don't trigger migration hooks
• branch.wipe.before and branch.wipe.after hooks respond to wipe, but they don't trigger migration hooks
• rebase and merge branch commands don't trigger branch hooks and instead trigger migration hooks (which is correct)
• I was able to chain simple commands directly in the gel.toml. I was able to see environment variables and run scripts in the root of the project directory as intended.

My original intent with the RFC was that any schema change to a consistent schema state should trigger migration hooks. That means all invocation of migrations should do it, but also any switches between branches (since each branch is in a consistent schema state, potentially different from previous branch). We can argue that the branch hooks can handle schema changes on their own and don't need to also trigger migration hooks. The original separation was to keep schema-related scripts connected to migration hooks and non-schema related config connected to branch and project init hooks. This kind of thing should follow some kind of intuition, although obviously the docs would disambiguate which hooks get triggered by which commands.

[vpetrovykh]

There's a hook behavior that was not specified in the RFC one way or another. When a hook fails, the operation aborts at that point. E.g. if the branch.switch.before hook fails, the current branch is not switched and the after hook is never called. This is probably reasonable, but we need to be careful when the hooks are called. gel branch switch newbranch -c will call the "before" hook after the new branch is created, but before it is switched. So the command executes partially.

[aljazerzen]

... will call the "before" hook after the new branch is created, but before it is switched

Oh. That's not good - it should be called before the branch is created. Execution of a command should be all or nothing.

---

Wait, are you saying that migration.apply.after should trigger after gel branch switch?