ref https://github.com/nim-lang/Nim/pull/25783
This pull request addresses an issue with addressability of tuple
elements of type `lent` or `var` in Nim, ensuring that expressions
involving these types are handled correctly during type changes. The
main changes introduce a check to prevent attempting to change the type
of tuple elements that are views (`var` or `lent`), and a new test is
added to verify the correct error is raised when trying to take the
address of such elements.
Type system and semantic analysis improvements:
* Added the `isViewTarget` template in `semexprs.nim` to check if a type
is a view (`var` or `lent`), and updated `changeType` to skip type
changes for tuple elements that are views. This prevents invalid
addressability operations on these types.
[[1]](diffhunk://#diff-539da3a63df08fa987f1b0c67d26cdc690753843d110b6bf0805a685eeaffd40R655-R657)
[[2]](diffhunk://#diff-539da3a63df08fa987f1b0c67d26cdc690753843d110b6bf0805a685eeaffd40R686-R693)
Testing:
* Added a new test `tlent_tuple_address.nim` to verify that attempting
to take the address of tuple elements of type `lent` correctly produces
an "expression has no address" error.
This fixes type resolution for `iterable[T]`.
I want to proceed with RFC
[#562](https://github.com/nim-lang/RFCs/issues/562) and this is the main
blocker for composability.
Fixes#22098 and, arguably, #19206
```nim
import std/strutils
template collect[T](it: iterable[T]): seq[T] =
block:
var res: seq[T] = @[]
for x in it:
res.add x
res
const text = "a b c d"
let words = text.split.collect()
doAssert words == @[ "a", "b", "c", "d" ]
```
In cases like `strutils.split`, where both proc and iterator overload
exists, the compiler resolves to the `func` overload causing a type
mismatch.
The old mode resolved `text.split` to `seq[string]` before the
surrounding `iterable[T]` requirement was applied, so the argument no
longer matched this template.
It should be noted that, compared to older sequtils templates,
composable chains based on `iterable[T]` require an iterator-producing
expression, e.g. `"foo".items.iterableTmpl()` rather than just
`"foo".iterableTmpl()`. This is actually desirable: it keeps the
iteration boundary explicit and makes iterable-driven templates
intentionally not directly interchangeable with older
untyped/loosely-typed templates like those in `sequtils`, whose internal
iterator setup we have zero control over (e.g. hard-coding adapters like
`items`).
Also, I noticed in `semstmts` that anonymous iterators are always
`closure`, which is not that surprising if you think about it, but still
I added a paragraph to the manual.
Regarding implementation:
From what I gathered, the root cause is that `semOpAux` eagerly
pre-types all arguments with plain flags before overload resolution
begins, so by the time `prepareOperand` processes `split` against the
`iterable[T]`, the wrong overload has already won.
The fix touches a few places:
- `prepareOperand` in `sigmatch.nim`:
When `formal.kind == tyIterable` and the argument was already typed as
something else, it's re-semchecked with the
`efPreferIteratorForIterable` flag. The recheck is limited to direct
calls (`a[0].kind in {nkIdent, nkAccQuoted, nkSym, nkOpenSym}`) to avoid
recursing through `semIndirectOp`/`semOpAux` again.
- `iteratorPreference` field `TCandidate`, checked before
`genericMatches` in `cmpCandidates`, gives the iterator overload a win
without touching the existing iterator heuristic used by `for` loops.
**Limitations:**
The implementation is still flag-driven rather than purely
formal-driven, so the behaviour is a bit too broad `efWantIterable` can
cause iterator results to be wrapped as `tyIterable` in
iterable-admitting contexts, not only when `iterable[T]` match is being
processed.
`iterable[T]` still does not accept closure iterator values such
as`iterator(): T {.closure.}`. It only matches the compiler's internal
`tyIterable`, not arbitrary iterator-typed values.
The existing iterator-preference heuristic is still in place, because
when I tried to remove it, some loosely-related regressions happened. In
particular, ordinary iterator-admitting contexts and iterator chains
still rely on early iterator preference during semchecking, before the
compiler has enough surrounding context to distinguish between
value/iterator producing overloads. Full heuristic removal would require
a broader refactor of dot-chain/intermediate-expression semchecking,
which is just too much for me ATM. This PR narrows only the
tyIterable-specific cases.
**Future work:**
Rework overload resolution to preserve additional information of
matching iterator overloads for calls up to the point where the
iterator-requiring context is established, to avoid re-sem in
`prepareOperand`.
Currently there's no good channel to store that information. Nodes can
get rewritten, TCandidate doesn't live long enough, storing in Context
or some side-table raises the question how to properly key that info.
fixes#25312
Tuple expressions `(a, b, c)` can be either types or values depending on
if their elements are typedescs or values, this is checked by checking
if the type of the element is `tyTypeDesc`. However when an
`skGenericParam` symbol is semchecked by `semSym` it is given its own
`tyGenericParam` type rather than a `tyTypeDesc` type, this seems to be
necessary for signatures to allow wildcard generic params passed to
static constrained generic params (tested in #25315). The reason
`semSym` is called is that `semGeneric` for generic invocations calls
`matches` which sems its arguments like normal expressions.
To deal with this, an expression of type `tyGenericParam` and with a
`skGenericParam` sym is allowed as a type in the tuple expression. A
problem is that this might consider a value with a wildcard generic
param type as a type. But this is a very niche problem, and I'm not sure
how to check for this. `skGenericParam` symbols stay as idents when
semchecked so it can't be checked that the node is an `skGenericParam`
symbol. It could be checked that it's an ident but I don't know how
robust this is. And maybe there is another way to refer to a wildcard
generic param type instead of just its symbol, i.e. another kind of
node.
This also makes #5647 finally work but a test case for that can be added
after.
TODO:
- [ ] test writing of .nif files
- [x] implement loading of fields in PType/PSym that might not have been
loaded
- [ ] implement interface logic
- [ ] implement pragma "replays"
- [ ] implement special logic for `converter`
- [ ] implement special logic for `method`
- [ ] test the logic holds up for `export`
- [ ] implement logic to free the memory of PSym/PType if memory
pressure is high
- [ ] implement logic to close memory mapped files if too many are open.
---------
Co-authored-by: demotomohiro <gpuppur@gmail.com>
Co-authored-by: ringabout <43030857+ringabout@users.noreply.github.com>
Co-authored-by: Jacek Sieka <arnetheduck@gmail.com>
fixes#23713
`linkTo` normally sets the sym of the type as well as the type of the
sym, but this is not wanted for custom pragmas as it would look up the
definition of the generic param and not the definition of its value. I
don't see a practical use for this either.
follows up #24871
For subscript assignments, if an overload of `[]=`/`{}=` is not found,
the LHS checks for overloads of `[]`/`{}` as a fallback, similar to what
field setters do since #24871. This is accomplished by just compiling
the LHS if the assignment overloads fail. This has the side effect that
the error messages are different now, instead of displaying the
overloads of `[]=`/`{}=` that did not match, it will display the ones
for `[]`/`{}` instead. This could be fixed by checking for `efLValue`
when giving the error messages for `[]`/`{}` but this is not done here.
The code for `[]` subscripts is a little different because of the
`mArrGet`/`mArrPut` overloads that always match. If the `mArrPut`
overload matches without a builtin subscript behavior for the LHS then
it calls `semAsgn` again with `mode = noOverloadedSubscript`. Before
this meant "fail to compile" but now it means "try to compile the LHS as
normal", in both cases the overloads of `[]=` are not considered again.
fixes#24947
When injectdestructors detects that a variable is a tuple unpacking temp
(i.e. it is an `skTemp`, is not a cursor, and has tuple type) it does
not generate a destructor for it and only generates sink/bit assignments
for its components. However the reason it does not generate a destructor
is that it expects it to be fully unpacked, this is true for unpackings
in for loops but not for tuple unpacking assignments which supports `_`
since #22537. Tuple unpacking definitions for `var`/`let`/`const` do not
generate `skTemp` and use the same symbol kind as the definition so they
did not have this problem.
To keep this compatible, the `_` parts of the tuple unpacking
assignments are now not ignored and unpacked into `let _ = ...`, which
generates its own destructor. Another option might be to use `skLet`
instead of `skTemp` but this might cause changes to behavior like
additional copies, I am not sure about this though.
refs https://forum.nim-lang.org/t/12785, refs #4711
The code was already there that when `propertyWriteAccess` returns `nil`
(i.e. cannot find a setter), `semAsgn` turns the [LHS into a call and
semchecks
it](1ef9a656d2/compiler/semexprs.nim (L1941-L1948)),
meaning if a setter cannot be found a getter will be assigned to
instead. However `propertyWriteAccess` never returned nil, because
`semOverloadedCallAnalyseEffects` was not called with `efNoUndeclared`
and so produced an error directly. So `efNoUndeclared` is passed to this
call so this code works as intended.
This fixes the issue described in #4711 which was closed because
subscripts do not have the same behavior implemented. However we can
implement this for subscripts as well (I have an implementation ready),
it just changes the error message from the failed overloads of `[]=` to
the failed overloads of `[]` for the LHS, which might be misleading but
is consistent with the error messages for any other assignment. I can do
this in this PR or another one.
To protect against crashes when this stops being experimental, in most
places handled the exact same as normal symchoices (not encountered in
typed ast)
fixes#24484, fixes#24672
When an array, set or tuple constructor has an element that resolves to
`tyFromExpr`, the type of the entire literal is now set to `tyFromExpr`
and the subsequent elements are not matched to any type.
The remaining expressions are still typed (a version of the PR before
this called `semGenericStmt` on them instead), however elements with int
literal types have their types set to `nil`, since generic instantiation
removes int literal types and the int literal type is required for
implicitly converting the int literal element to the set type. Tuples
should not really need this but it is done for them anyway in case it
messes up some type inference
---------
Co-authored-by: Andreas Rumpf <rumpf_a@web.de>
ref https://github.com/nim-lang/Nim/issues/24764
To keep destructors injected consistently, we need to transform `mAsgn`
properly into `nkSinkAsgn` and `nkAsgn`. This PR is the first step
towards overhauling hook injections.
In this PR, hooks (except mAsgn) are treated consistently whether it is
resolved in matching or instantiated by sempass2. It also fixes a
spelling `=wasMoved` to its normalized version, which caused no
replacing generic hook calls with lifted hook calls.
fixes#24755, refs #24710
Instead of using the node from `indexTypesMatch` which inserts a hidden
conv node, just change the type of the node back to the old type
directly
fixes#24698
The same aim as #24224 but for tuple constructors. The difference here
is that the type of a tuple constructor is always going to be valid
unlike array constructors which can have `seq` etc types, so we can just
generate a conversion again. If the conversion fails, it is ignored
similar to #24611, this is to protect against modified typed nodes in
macros.
Also #24611 was only adapted to `semTupleFieldsConstr` and not
`semTuplePositionsConstr`, this is now fixed.
fixes#24609
A tuple may have an incompatible expected type if there is a converter
match to it. So the compiler should not error when trying to match the
individual elements in the constructor to the elements of the expected
tuple type, this will be checked when the tuple is entirely constructed
anyway.
fixes#24492
Kind of a goofy way of doing this, but we count how many derefs were
used for the first parameter before calling `builtinFieldAccess`, then
count after, and if there are more now than before, we remove the added
derefs. I thought maybe getting rid of #18298 would simplify it but
maybe this would still be the best way.
For better encapsulation we could make `dotTransformation` take an
`nOrig` param instead but this would be less efficient since it would
need a copy, though `semAsgn` already makes one.
fixes issue described in https://forum.nim-lang.org/t/12579
In #24065 explicit generic parameter matching was made to fail matches
on arguments with unresolved types in generic contexts (the sigmatch
diff, following #24010), similar to what is done for regular calls since
#22029. However unlike regular calls, a failed match in a generic
context for a standalone explicit generic instantiation did not convert
the expression into one with `tyFromExpr` type, which means it would
error immediately given any unresolved parameter. This is now done to
fix the issue.
For explicit generic instantiations on single non-overloaded symbols, a
successful match is still instantiated. For multiple overloads (i.e.
symchoice), if any of the overloads fail the match, the entire
expression is considered untyped and any instantiations are not used, so
as to not void overloads that would match later. This means even
symchoices without unresolved arguments aren't instantiated, which may
be too restrictive, but it could also be too lenient and we might need
to make symchoice instantiations always untyped. The behavior for
symchoice is not sound anyway given it causes #9997 so this is something
to consider for a redesign.
Diff follows #24276.
refs #8064, refs #24010
Error messages for standalone explicit generic instantiations are
revamped. Failing standalone explicit generic instantiations now only
error after overloading has finished and resolved to the default `[]`
magic (this means `[]` can now be overloaded for procs but this isn't
necessarily intentional, in #24010 it was documented that it isn't
possible). The error messages for failed instantiations are also no
longer a simple `cannot instantiate: foo` message, instead they now give
the same type mismatch error message as overloads with mismatching
explicit generic parameters.
This is now possible due to the changes in #24010 that delegate all
explicit generic proc instantiations to overload resolution. Old code
that worked around this is now removed. `maybeInstantiateGeneric` could
maybe also be removed in favor of just `explicitGenericSym`, the `result
== n` case is due to `explicitGenericInstError` which is only for niche
cases.
Also, to cause "ambiguous identifier" error messages when the explicit
instantiation is a symchoice and the expression context doesn't allow
symchoices, we semcheck the sym/symchoice created by
`explicitGenericSym` with the given expression flags.
#8064 isn't entirely fixed because the error message is still misleading
for the original case which does `values[1]`, as a consequence of
#12664.
split from #24198
This is a required refactor for the only good solution I've been able to
think of for #4858 etc. Explanation:
---
`sigmatch` currently [disables
bindings](d6a71a1067/compiler/sigmatch.nim (L1956))
(except for binding to other generic parameters) when matching against
constraints of generic parameters. This is so when the constraint is a
general metatype like `seq`, the type matching will not treat all
following uses of `seq` as the type matched against that generic
parameter.
However to solve #4858 etc we need to bind `or` types with a conversion
match to the type they are supposed to be converted to (i.e. matching
`int literal(123)` against `int8 | int16` should bind `int8`[^1], not
`int`). The generic parameter constraint binding needs some way to keep
track of this so that matching `int literal(123)` against `T: int8 |
int16` also binds `T` to `int8`[^1].
The only good way to do this IMO is to generate a new "binding context"
when matching against constraints, then binding the generic param to
what the constraint was bound to in that context (in #24198 this is
restricted to just `or` types & concrete types with convertible matches,
it doesn't work in general).
---
`semtypinst` already does something similar for bindings of generic
invocations using `LayeredIdTable`, so `LayeredIdTable` is now split
into its own module and used in `sigmatch` for type bindings as well,
rather than a single-layer `TypeMapping`. Other modules which act on
`sigmatch`'s binding map are also updated to use this type instead.
The type is also made into an `object` type rather than a `ref object`
to reduce the pointer indirection when embedding it inside
`TCandidate`/`TReplTypeVars`, but only on arc/orc since there are some
weird aliasing bugs on refc/markAndSweep that cause a segfault when
setting a layer to its previous layer. If we want we can also just
remove the conditional compilation altogether and always use `ref
object` at the cost of some performance.
[^1]: `int8` binding here and not `int16` might seem weird, since they
match equally well. But we need to resolve the ambiguity here, in #24012
I tested disallowing ambiguities like this and it broke many packages
that tries to match int literals to things like `int16 | uint16` or
`int8 | int16`. Instead of making these packages stop working I think
it's better we resolve the ambiguity with a rule like "the earliest `or`
branch with the best match, matches". This is the rule used in #24198.
fixes#23010, split from #24195
When resemming bracket nodes, the compiler currently unconditionally
makes a new node with an array type based on the node. However the VM
can generate bracket nodes with `seq` types, which this erases. To fix
this, if a bracket node already has a type, we still resem the bracket
node, but don't construct a new type for it, instead using the type of
the original node.
A version of this was rejected that didn't resem the node at all if it
was typed, but I can't find it. The difference with this one is that the
individual elements are still resemmed.
This should fix the break caused by #24184 so we could redo it after
this PR but it might still have issues, not to mention the related
pre-existing issues like #22793, #12559 etc.
fixes#24179
The original fix made it so calls to `skError`/`skUnknown` (in this case
`->`, for some reason `sugar` couldn't be imported) returned an error
node, however this breaks tsug_accquote for some reason I don't
understand (it even parses as `tsug_accquote.discard`) so I've just
added a guard based on the stacktrace.
fixes#24164, regression from #20091
The expression `nil` as the default value of template parameter `x:
untyped` is typechecked with expected type `untyped` since #20091. The
expected type is checked if it matches the `nil` literal with a match
better than a subtype match, and the type is set to it if it does.
However `untyped` matches with a generic match which is better, so the
`nil` literal has type `untyped`. This breaks type matching for the
literal. So if the expected type is `untyped` or `typed`, it is now
ignored and the `nil` literal just has the `nil` type.
alternative to #24101#23892 changed the opensym experimental switch so that it has to be
enabled in the context of the generic/template declarations capturing
the symbols, not the context of the instantiation of the
generics/templates. This was to be in line with where the compiler gives
the warnings and changes behavior in a potentially breaking way.
However `results` [depends on the old
behavior](71d404b314/results.nim (L1428)),
so that the callers of the macros provided by results always take
advantage of the opensym behavior. To accomodate this, we change the
behavior of the old experimental option that `results` uses,
`genericsOpenSym`, so that ignores the information of whether or not
symbols are intentionally opened and always gives the opensym behavior
as long as it's enabled at instantiation time. This should keep
`results` working as is. However this differs from the normal opensym
switch in that it doesn't generate `nnkOpenSym`.
Before it was just a generics-only version of `openSym` along with
`templateOpenSym` which was only for templates. So `templateOpenSym` is
removed along with this change, but no one appears to have used it.
fixes regression remaining after #24092
In #24092 `prepareNode` was updated so it wouldn't try to instantiate
generic type symbols (like `Generic` when `type Generic[T] = object`,
and `prepareNode` is what `tyFromExpr` uses in most of the compiler. An
exception is in sigmatch, which is now changed to use `prepareNode` to
make generic type symbols work in the same way as usual. However this
requires another change to work:
Dot fields and matches to `typedesc` on generic types generate
`tyFromExpr` in generic contexts since #24005, including generic type
symbols. But this means when we try to instantiate the `tyFromExpr` in
sigmatch, which increases `c.inGenericContext` for potentially remaining
unresolved expressions, dotcalls stay as `tyFromExpr` and so never
match. To fix this, we change the "generic type" check in dot fields and
`typedesc` matching to an "unresolved type" check which excludes generic
body types; and for generic body types, we only generate `tyFromExpr` if
the dot field is a generic parameter of the generic type (so that it
gets resolved only at instantiation).
Notes for the future:
* Sigmatch shouldn't have to `inc c.inGenericContext`, if a `tyFromExpr`
can't instantiate it's fine if we just fail the match (i.e. redirect the
instantiation errors from `semtypinst` to a match failure). Then again
maybe this is the best way to check for inability to instantiate.
* The `elif c.inGenericContext > 0 and t.containsUnresolvedType` check
in dotfields could maybe be simplified to just checking for `tyFromExpr`
and `tyGenericParam`, but I don't know if this is an exhaustive list.
fixes#23506#24065 broke compilation of template parameter default values that
depended on other template parameters. But this was never implemented
anyway, actually attempting to use those default values breaks the
compiler as in #23506. So these are now implemented as well as fixing
the regression.
First, if a default value expression uses any unresolved arguments
(generic or normal template parameters) in a template header, we leave
it untyped, instead of applying the generic typechecking (fixing the
regression). Then, just before the body of the template is about to be
explored, the default value expressions are handled in the same manner
as the body as well. This captures symbols including the parameters, so
the expression is checked again if it contains a parameter symbol, and
marked with `nfDefaultRefsParam` if it does (as an optimization to not
check it later).
Then when the template is being evaluated, when substituting a
parameter, if we try to substitute with a node marked
`nfDefaultRefsParam`, we also evaluate it as we would the template body
instead of passing it as just a copy (the reason why it never worked
before). This way we save time if a default value doesn't refer to
another parameter and could just be copied regardless.
fixes CI, refs #24066, refs #24065
The combination of #24065 and #24066 caused a CI failure where a `when`
branch that was never compiled gave an undeclared identifier error. This
is because the `when` branch was being semchecked with `semGenericStmt`
without `withinMixin`, which is the flag `semgnrc` normally gives to
`when` branch bodies. To fix this, just pass the whole `when` stmt to
`semGenericStmt` rather than the individual blocks.
The alternative would be to just replace the calls to `semGenericStmt`
with a new proc that does the same thing, just with the flags
`{withinMixin}`.
I noticed after #24005 the auto-reported boot time in PRs increased from
around 8 seconds to 8.8 seconds, but I wasn't sure what could cause a
performance problem that made the compiler itself compile slower, most
of the changes were related to `static` which the compiler code doesn't
use too often. So I figured it was unrelated.
However there is still a performance problem with the changes to
`tryReadingGenericParam`. If an expression like `a.b` doesn't match any
of the default dot field behavior (for example, is actually a call
`b(a)`), the compiler does a final check to see if `b` is a generic
parameter of `a`. Since #24005, if the type of `a` is not
`tyGenericInst` or an old concept type, the compiler does a full
traversal of the type of `a` to see if it contains a generic type, only
then checking for `c.inGenericContext > 0` to not return `nil`. This
happens on *every* dot call.
Instead, we now check for `c.inGenericContext > 0` first, only then
checking if it contains a generic type, saving performance by virtue of
`c.inGenericContext > 0` being both cheap and less commonly true. The
`containsGenericType` could also be swapped out for more generic type
kind checks, but I think this is incorrect even if it might pass CI.
fixes#22342, fixes#22607
Another followup of #22029, `when` expressions in general in generic
type bodies now behave like `nkRecWhen` does since #24042, leaving them
as `tyFromExpr` if a condition is uncertain. The tests for the issues
were originally added but left disabled in #24005.
fixes#16700, fixes#20916, refs #24010
Fixes the instantiation issues for proc param default values encountered
in #24010 by:
1. semchecking generic default param values with `inGenericContext` for
#22029 and followups to apply (the bigger change in semtypes),
2. rejecting explicit generic instantiations with unresolved generic
types inside `inGenericContext` (sigmatch change),
3. instantiating the default param values using `prepareNode` rather
than an insufficient manual method (the bigger change in seminst).
This had an important side effect of references to other parameters not
working since they would be resolved as a symbol belonging to the
uninstantiated original generic proc rather than the later instantiated
proc. There is a more radical way to fix this which is generating ident
nodes with `tyFromExpr` in specifically this context, but instead we
just count them as belonging to the same proc in
`hoistParamsUsedInDefault`.
Other minor bugfixes:
* To make the error message in t20883 make sense, we now give a "cannot
instantiate" error when trying to instantiate a proc generic param with
`tyFromExpr`.
* Object constructors as default param values generated default values
of private fields going through `evalConstExpr` more than once, but the
VM doesn't mark the object fields as `nfSkipFieldChecking` which gives a
"cannot access field" error. So the VM now marks object fields it
generates as `nfSkipFieldChecking`. Not sure if this affects VM
performance, don't see why it would.
* The nkRecWhen changes in #24042 didn't cover the case where all
conditions are constantly false correctly, this is fixed with a minor
change. This isn't needed for this PR now but I encountered it after
forgetting to `dec c.inGenericContext`.
The magic `mArrGet`/`mArrPut` subscript overloads always match, so if a
subscript doesn't match any other subscript overloads and isn't a
regular language-handled subscript, it creates a fake overload mismatch
error in `semArrGet` that doesn't have any information (gives stuff like
"first mismatch at index: 0" for every single mismatch). Instead of
generating the fake mismatches, we only generate the fake mismatch for
`mArrGet`/`mArrPut`, and process every overload except them as a real
call and get the errors from there.
refs #21682, refs #24038
The `rangeBase` typetrait added in #21682 which gives the base type of a
range type is now added publicly to `typetraits`. Previously it was only
privately used in `repr_v2` and in `enumutils` since #24052
(coincidentally I didn't see this until now). This is part of an effort
to make range types easier to work with in generics, as mentioned in
#24038. Its use combined with #24037 is also tested.
The condition for the "enum to enum conversion" warning is now also
restricted to conversions between different enum base types, i.e.
conversion between an enum type and a range type of itself doesn't give
a warning. I put this in this PR since the test gave the warning and so
works as a regression test.
fixes#16376
The way the compiler handled generic proc instantiations in calls (like
`foo[int](...)`) up to this point was to instantiate `foo[int]`, create
a symbol for the instantiated proc (or a symchoice for multiple procs
excluding ones with mismatching generic param counts), then perform
overload resolution on this symbol/symchoice. The exception to this was
when the called symbol was already a symchoice node, in which case it
wasn't instantiated and overloading was called directly ([these
lines](b7b1313d21/compiler/semexprs.nim (L3366-L3371))).
This has several problems:
* Templates and macros can't create instantiated symbols, so they
couldn't participate in overloaded explicit generic instantiations,
causing the issue #16376.
* Every single proc that can be instantiated with the given generic
params is fully instantiated including the body. #9997 is about this but
isn't fixed here since the instantiation isn't in a call.
The way overload resolution handles explicit instantiations by itself is
also buggy:
* It doesn't check constraints.
* It allows only partially providing the generic parameters, which makes
sense for implicit generics, but can cause ambiguity in overloading.
Here is how this PR deals with these problems:
* Overload resolution now always handles explicit generic instantiations
in calls, in `initCandidate`, as long as the symbol resolves to a
routine symbol.
* Overload resolution now checks the generic params for constraints and
correct parameter count (ignoring implicit params). If these don't
match, the entire overload is considered as not matching and not
instantiated.
* Special error messages are added for mismatching/missing/extra generic
params. This is almost all of the diff in `semcall`.
* Procs with matching generic parameters now instantiate only the type
of the signature in overload resolution, not the proc itself, which also
works for templates and macros.
Unfortunately we can't entirely remove instantiations because overload
resolution can't handle some cases with uninstantiated types even though
it's resolved in the binding (see the last 2 blocks in
`texplicitgenerics`). There are also some instantiation issues with
default params that #24005 didn't fix but I didn't want this to become
the 3rd huge generics PR in a row so I didn't dive too deep into trying
to fix them. There is still a minor instantiation fix in `semtypinst`
though for subscripts in calls.
Additional changes:
* Overloading of `[]` wasn't documented properly, it somewhat is now
because we need to mention the limitation that it can't be done for
generic procs/types.
* Tests can now enable the new type mismatch errors with just
`-d:testsConciseTypeMismatch` in the command.
Package PRs:
- using fork for now:
[combparser](https://github.com/PMunch/combparser/pull/7) (partial
generic instantiation)
- merged: [cligen](https://github.com/c-blake/cligen/pull/233) (partial
generic instantiation but non-overloaded + template)
- merged: [neo](https://github.com/andreaferretti/neo/pull/56) (trying
to instantiate template with no generic param)
