macro pragmas: support template and const defs #388
Conversation
saem
force-pushed
the
saem-support-macro-pragma-on-template
branch
from
db444ce
to
0a4dbe7
Compare
|
bors r+ |
388: macro pragmas: support template and const defs r=saem a=saem
## Summary:
* constdef annotation handling same as identdefs in let or var
* annotation processing on template definitions like other routine kinds
## Details:
* reenabled `tests/pragmas/tpragmas_misc`
* updated manual for macros pragmas
* `semstmts.semConst` reworked to be like `semstmts.semLetOrVar`
* constdef annotation handling same as identdefs in let or var
* `semLetOrVarAnnotation` works on const defs, and renamed
* added tracing to `semTemplateDef`
* added tests for let/var section processing
Mostly a port of nim-lang/Nim#19406
Co-authored-by: saem <saemghani+github@gmail.com>
| if def.kind != nkNilLit: | ||
| # xxx: this seems wrong, instead of guarding on nil, maybe it should be | ||
| # inferred from `typ`? then we still do the rest of the checks? |
There was a problem hiding this comment.
The nkNilLit check here is done in order to support nil ref, ptr and pointer values for consts (they would be rejected by typeAllowed otherwise).
There was a problem hiding this comment.
Yeah, but that sounds like a bug with typeAllowed that we're working around here?
There was a problem hiding this comment.
Well, typeAllowed only looks at the type, so I wouldn't consider it a bug.
Thinking about this a bit more, we don't want to disallow ref and ptr in general, but only non-nil values of these types. With the current implementation,
type A = object
x: ref int
const c = A()is rejected, even though it'd work fine.
Rejecting non-nil values for ptr, ref, and closure environments is actually already implemented (the rsemVmUnsupportedNonNil report; see deserializeRef and deserialize in vmcompilerserdes.nim), so the nkNilLit guard here could be removed (after adjusting typeAllowed).
typeAllowed is still needed here, in order to reject objects using inheritance (the back-ends don't support those yet if I remember correctly).
I do think it's better to do this in a separate PR however.
There was a problem hiding this comment.
Totally agree on the separate PR.
As for whether it's a bug or not. typeAllowed does know it's a const/let/var. It's also only receiving typ, and not def. So the only pathway for def to influence typeAllowed is the earlier derivation of typ from def. In which case, shouldn't typ be correctly constructed such that typeAllowed doesn't reject it?
Correctly constructed meaning, given:
const foo: SomeRefType = nil
# or
const foo: SomeRefType = evalsToNil()
even if def is a nil it should make no difference as typ should be correctly derived. Unless, typ is not being derived correctly from def and/or we're missing a user specified type expession and we have a pure nil value (kinda like a void pointer).
Maybe I'm totally misunderstanding it?
|
bors cancel |
|
Canceled. |
saem
force-pushed
the
saem-support-macro-pragma-on-template
branch
4 times, most recently
from
46cbf13
to
45ce390
Compare
|
Squashed with all the latest. |
|
bors r+ |
|
👎 Rejected by PR status |
|
bors r+ |
|
👎 Rejected by PR status |
* constdef annotation handling same as identdefs in let or var * annotation processing on template definitions like other routine kinds Details: * reenabled `tests/pragmas/tpragmas_misc` * updated manual for macros pragmas * `semstmts.semConst` reworked to be like `semstmts.semLetOrVar` * constdef annotation handling same as identdefs in let or var * `semLetOrVarAnnotation` works on const defs, and renamed * added tracing to `semTemplateDef` * added tests for let/var section processing Mostly a port of nim-lang/Nim#19406
saem
force-pushed
the
saem-support-macro-pragma-on-template
branch
from
45ce390
to
a0f8e2a
Compare
|
Not sure what exactly broke reproducible builds, have to figure out how to figure that out. 😓 But in the meantime rebased on the latest devel as it contains some workflow updates and perhaps that's might address this. 🤞🏽 |
|
bors r+ |
|
Build succeeded: |
Summary:
Details:
tests/pragmas/tpragmas_miscsemstmts.semConstreworked to be likesemstmts.semLetOrVarsemLetOrVarAnnotationworks on const defs, and renamedsemTemplateDefMostly a port of nim-lang/Nim#19406