explicit_object_extensions provisional -- SIMICS-23313#399
explicit_object_extensions provisional -- SIMICS-23313#399lwaern-intel wants to merge 1 commit intointel:mainfrom
explicit_object_extensions provisional -- SIMICS-23313#399Conversation
lwaern-intel
force-pushed
the
lw/explicit-object-decls
branch
from
bcd632e to
aa03516
Compare
| # extension status: | ||
| # None -> dual extension and decl | ||
| # True -> extension | ||
| # False -> explicit decl |
There was a problem hiding this comment.
This ternary rep is pretty damn ugly. @mandolaerik does this deserve the use of enum (which'd be a fresh DMLC dependency?) Or is there a third approach perhaps I haven't thought of?
There was a problem hiding this comment.
The current rep is fine, but I also wouldn't object to enum. The question it poses to me is rather whether it makes sense to handle subobj merge in this incremental pairwise fashion -- I think this logic would be much easier to express if we just merged all object defs at once.
IIRC the pairwise thing is there for a historical reason: before dml 1.4, templates would consist of pre-merged subobj defs, so a template used many times did not have to merge defs again.
It would probably make sense to refactor all of merge_subobj_defs, but no need to do within this PR.
| fmt = ("object '%s' not declared elsewhere." | ||
| " To declare and define a new object, omit 'in'.") | ||
|
|
||
| class EMULTIOBJDECL(DMLError): |
There was a problem hiding this comment.
I struggled a bit with naming of these error tags, but ultimately it doesn't matter much.
|
PR Verification: ✅ success |
|
@mandolaerik ping |
| 'object : REGISTER objident array_list sizespec offsetspec maybe_istemplate object_spec' | ||
| t[0] = ast.object_(site(t), t[2], 'register', t[3], | ||
| t[4] + t[5] + t[6] + t[7]) | ||
| 'object : maybe_extension REGISTER objident array_list sizespec offsetspec maybe_istemplate object_spec' |
There was a problem hiding this comment.
do we really want to allow in register x size 4 @ 4711 { } ? all the special syntax is a clear declaration marker to me.
I suggest in KIND NAME [INDICES] { ... } as the only allowed form, analogous to the grammar for in each xyz { ... }
BTW: should we forbid the [_ < ...] index forms outside extensions when the provisional is active?
There was a problem hiding this comment.
do we really want to allow in register x size 4 @ 4711 { } ?
I reference this in the JIRA issue. I intentionally made this feature as consistent as possible, and the questions of what we should or should not allow a separate issue -- because the answers are not obvious, (in particular, although I previously agreed with forbidding [_ < ...] as a decl, upon reevaluation it's less obvious to me that such a decl would be nonsensical.)
In other words: i don't care when it comes to this PR. To answer these questions, we need deeper discussion and analysis, and this PR has no need for that since this is an unstable provisional. Besides, the consistent design is a virtue in its own right, and answers about it vs. any ad-hoc restrictions we make could perhaps be enlightened by Gustav's use of it.
| @feature | ||
| class explicit_object_extensions(ProvisionalFeature): | ||
| '''<a id="explicit_object_extensions"/> | ||
| This feature extends the DML syntax for object declarations to distinguish |
There was a problem hiding this comment.
newline after <a, otherwise markdown is messed up in the first paragraph.
There was a problem hiding this comment.
wait-- you mean
<a
id="explicit_object_extensions"/>
? That's messed up, man...
There was a problem hiding this comment.
wait i can't see that reflected for explicit_param_decls. Does that mean its docs are messed up?
There was a problem hiding this comment.
... in fact I can't see a <a-newline pattern being used anywhere?
There was a problem hiding this comment.
no, I meant to write:
<a id="foo"/>
blah `blah`
instead of:
<a id="foo"/>
blah `blah`
The first one is rendered like this:
blah blah
The second one is rendered like this:
blah `blah`and don't ask me why everything is underlined now
| " To declare and define a new object, omit 'in'.") | ||
|
|
||
| class EMULTIOBJDECL(DMLError): | ||
| """When the `explicit_object_extensions` provisional feature is enabled, |
|
|
||
|
|
||
| class EEXTENSION(DMLError): | ||
| """When the `explict_object_extensions` provisional feature is enabled, |
3 participants
No description provided.