stage2: astgen for try

[g-w1]

I used orelseCatchExpr as inspiration for most of the tryExpr function. This is my first astgen expr, so please tell me if something is not idomatic.

I also moved the -femit-zir code higher up in the updateModule function so that even if sema fails, you still get the ir in $file.zir.

I couldn't find a way to test this.

[LemonBoy]

See #3461 for a stage1 bug related to how try is desugared.
See #7541 for another stage1 bug that's related to the very same thing, this time caused by the explosion of try in many simple (and unlinked) IR pieces that may get constant-folded separately,

[g-w1]

Wow, everything is coming together. I was just thinking about how to implement inferred error sets yesterday. Both of these issues stem from that. The call that checks if the errorset matches the return type is this: https://github.com/ziglang/zig/pull/7606/files#diff-2600885f5a9e88bc263d2b47daab6d1c33418966af8f42b26c188597c108201bR930-R940 I think since I will probably also be implementing inferred error sets soon, I will keep this in the back of my head. If these bugs were to happen, they would not happen in astgen for try but in the coercion code and semantic analysis code for inferred/regular error sets.

[LemonBoy]

If these bugs were to happen, they would not happen in astgen for try but in the coercion code and semantic analysis code for inferred/regular error sets.

We'll see.

[andrewrk]

I'd like to get orelse/catch hooked up all the way through the compiler with test coverage before adding more astgen that uses similar logic, to make sure we are going in the right direction here.

[g-w1]

I would too, #7628 will do that, then we can look at this.

…

On Mon Jan 18, 2021 at 7:29 AM EST, Andrew Kelley wrote: @andrewrk requested changes on this pull request. I'd like to get orelse/catch hooked up all the way through the compiler with test coverage before adding more astgen that uses similar logic, to make sure we are going in the right direction here. -- You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub: #7606 (review)

[andrewrk]

This PR will have to be re-done now that #7847 is landed.