This PR speeds up the calculation of the power of a complex number when
the exponent is 2.0 or 0.5 (i.e the square and the square root of a
complex number). These are probably two of (if not) the most common
exponents. The speed up that is achieved according to my measurements
(using the timeit library) when the exponent is set to 2.0 or 0.5 is >
x7, while there is no measurable difference when using other exponents.
For the record, this is the function I used to mesure the performance:
```nim
import std/complex
import timeit
proc calculcatePows(v: seq[Complex], factor: Complex): seq[Complex] {.noinit, discardable.} =
result = newSeq[Complex](v.len)
for n in 0 ..< v.len:
result[n] = pow(v[n], factor)
let v: seq[Complex64] = collect:
for n in 0 ..< 1000:
complex(float(n))
echo timeGo(calculcatePows(v, complex(1.5)))
echo timeGo(calculcatePows(v, complex(0.5)))
echo timeGo(calculcatePows(v, complex(2.0)))
```
Which with the original code got:
> [177μs 857.03ns] ± [1μs 234.85ns] per loop (mean ± std. dev. of 7
runs, 1000 loops each)
> [128μs 217.92ns] ± [1μs 630.93ns] per loop (mean ± std. dev. of 7
runs, 1000 loops each)
> [136μs 220.16ns] ± [3μs 475.56ns] per loop (mean ± std. dev. of 7
runs, 1000 loops each)
While with the improved code got:
> [176μs 884.30ns] ± [1μs 307.30ns] per loop (mean ± std. dev. of 7
runs, 1000 loops each)
> [23μs 160.79ns] ± [340.18ns] per loop (mean ± std. dev. of 7 runs,
10000 loops each)
> [19μs 93.29ns] ± [1μs 128.92ns] per loop (mean ± std. dev. of 7 runs,
10000 loops each)
That is, the new optimized path is 5.6 (23 vs 128 us per loop) to 7.16
times faster (19 vs 136 us per loop), while the non-optimized path takes
the same time as the original code.
Makes docs default to using browser settings instead of light mode
This should fix#16515 since it doesn't require the browser to run the
JS to set the default
Also means that dark mode can be used without JS if the browser is
configured to default to dark mode
fixes#18125
Previously a tuple type like `(T, int)` would match an expected tuple
type `(U, int)` if `T` is a subtype of `U`. This is wrong since the
codegen does not handle type conversions of individual tuple elements in
a type conversion of an entire tuple. For this reason the compiler
already does not accept `(float, int)` for a matched type `(int, int)`,
however the code that checked for which relations are unacceptable
checked for `< isSubtype` rather than `<= isSubtype`, so subtypes were
not included in the unacceptable relations.
Update: Now only considered unacceptable when inheritance is used, as in
[`paramTypesMatch`](3379d26629/compiler/sigmatch.nim (L2252-L2254)).
Ideally subtype relations that don't need conversions, like `nil`,
`seq[empty]`, `range[0..5]` etc would be their own relation
`isConcreteSubtype` (which would also allow us to differentiate with
`openArray[T]`), but this is too big of a refactor for now.
To compensate for this making things like `let x: (Parent, int) =
(Child(), 0)` not compile (they would crash codegen before anyway but
should still work in principle), type inference for tuple constructors
is updated such that they call `fitNode` on the fields and their
expected types, so a type conversion is generated for the individual
subtype element.
fixes#17163, refs #23204
Types that aren't `tyRange` and are bigger than 16 bits, so `int32`,
`uint64`, `int` etc, are disallowed as array index range types.
`tyRange` is excluded because the max array size is backend independent
(except for the specific size of `high(uint64)` which crashes the
compiler) and so there should still be an escape hatch for people who
want bigger arrays.
This PR modifies `specializeResetT` so that it generates the proper
reset code for range types. I've tested it in the examples for issues
#23214 and #22462 as well as our codebase, and it seems to fix the
issues I had been experiencing.
fixes#23200, fixes#18866#21065 made it so `auto` proc return types remained as `tyAnything` and
not turned to `tyUntyped`. This had the side effect that anything
previously bound to `tyAnything` in the proc type match was then bound
to the proc return type, which is wrong since we don't know the proc
return type even if we know the expected parameter types (`tyUntyped`
also [does not care about its previous bindings in
`typeRel`](ab4278d217/compiler/sigmatch.nim (L1059-L1061))
maybe for this reason).
Now we mark `tyAnything` return types for routines as `tfRetType` [as
done for other meta return
types](18b5fb256d/compiler/semtypes.nim (L1451)),
and ignore bindings to `tyAnything` + `tfRetType` types in `semtypinst`.
On top of this, we reset the type relation in `paramTypesMatch` only
after creating the instantiation (instead of trusting
`isInferred`/`isInferredConvertible` before creating the instantiation),
using the same mechanism that `isBothMetaConvertible` uses.
This fixes the issues as well as making the disabled t15386_2 test
introduced in #21065 work. As seen in the changes for the other tests,
the error messages give an obscure `proc (a: GenericParam): auto` now,
but it does give the correct error that the overload doesn't match
instead of matching the overload pre-emptively and expecting a specific
return type.
tsugar had to be changed due to #16906, which is the problem where
`void` is not inferred in the case where `result` was never touched.
fixes#23186
As explained in #23186, generics can transform `genericProc[int]` into a
call `` `[]`(genericProc, int) `` which causes a problem when
`genericProc` is resemmed, since it is not a resolved generic proc. `[]`
needs unresolved generic procs since `mArrGet` also handles explicit
generic instantiations, so delay the resolved generic proc check to
`semFinishOperands` which is intentionally not called for `mArrGet`.
The root issue for
[t6137](https://github.com/nim-lang/Nim/blob/devel/tests/generics/t6137.nim)
is also fixed (because this change breaks it otherwise), the compiler
doesn't consider the possibility that an assigned generic param can be
an unresolved static value (note the line `if t.kind == tyStatic: s.ast
= t.n` below the change in sigmatch), now it properly errors that it
couldn't instantiate it as it would for a type param. ~~The change in
semtypinst is just for symmetry with the code above it which also gives
a `cannot instantiate` error, it may or may not be necessary/correct.~~
Now removed, I don't think it was correct.
Still possible that this has unintended consequences.
Nim manual says:
> When using the Cpp backend, params marked as byref will translate to
cpp references `&`
But how `byref` pragma translate to depends on whether it is used with
`importc` or `importcpp`.
When `byref` pragma used with `importc` types and compiled with the Cpp
backend, it is not traslated to cpp reference `&`.
---------
Co-authored-by: Andreas Rumpf <rumpf_a@web.de>
- Clarified the implications of order of operation.
- Mentioned overlapping isn't handled
- Added the runnableExamples block
Fixes#23160, which supposedly should have been fixed in an earlier PR
#23022, but the wording was still not clear enough to my liking, which
the raised issue kind of confirms.
fixes#23177
`changeType` doesn't perform range checks to see if the expression fits
the new type [if the old type is the same as the new
type](62d8ca4306/compiler/semexprs.nim (L633)).
For `nkIntLit`, we previously set the type to the concrete base of the
expected type first, then call `changeType`, which works for things like
range types but not bare types of smaller bit size like `int8`. Now we
don't set the type (so the type is nil), and `changeType` performs the
range check when the type is unset (nil).
fixes#22775
It's pre-existing that [`prepareOperand` doesn't typecheck expressions
which have
types](a4f3bf3742/compiler/sigmatch.nim (L2444)).
Templates can take typed subscript expressions, transform them into
calls to `[]`, and then have this `[]` not be resolved later if the
expression is nested inside of a call argument, which leaks an untyped
expression past semantic analysis. To prevent this, don't transform any
typed subscript expressions into calls to `[]` in templates. Ditto for
curly subscripts (with `{}`) and assignments to subscripts and curly
subscripts (with `[]=` and `{}=`).
This is in reference to a [feature
request](https://github.com/nim-lang/Nim/issues/22142) that I posted.
I'm making this PR to demonstrate the suggested change and expect that
this should be scrutinized
---------
Co-authored-by: Bung <crc32@qq.com>
Co-authored-by: Andreas Rumpf <rumpf_a@web.de>
This PR modernises the NEP1 style guide to prefer hanging indent over
vertial alignment for long code statements while still allowing
alignment in legacy code.
The change is based on research and study of existing style guides for
both braced and indented languages that have seen wide adoption as well
as working with a large Nim codebase with several teams touching the
same code regularly.
The research was done as part of due diligence leading up to
[nph](https://github.com/arnetheduck/nph) which uses this style
throughout.
There are several reasons why hanging indent works well for
collaboration, good code practices and modern Nim features:
* as NEP1 itself points out, alignment causes unnecessary friction when
refactoring, adding/removing items to lists and otherwise improving code
style or due to the need for realignment - the new recommendation aligns
NEP1 with itself
* When collaborating, alignment leads to unnecessary git conflicts and
blame changes - with hanging indent, such conflicts are minimised.
* Vertical alignment pushes much of the code to the right where often
there is little space - when using modern features such as generics
where types may be composed of several (descriptively named) components,
there is simply no more room for parameters or comments
* The space to the left of the alignemnt cannot productively be used for
anything (unlike on the right, where comments may be placed)
* Double hanging indent maintaines visual separation between parameters
/ condition and the body that follows.
This may seem like a drastic change, but in reality, it is not:
* the most popular editor for Nim (vscode) already promotes this style
by default (if you press enter after `(`, it will jump to an indent on
the next line)
* although orthogonal to these changes, tools such as `nph` can be used
to reformat existing code should this be desired - when done in a single
commit, `git blame` is not lost and neither are exsting PRs (they can
simply be reformatted deterministically) - `nph` is also integrated with
vscode.
* It only affects long lines - ie most code remains unchanged
Examples of vertical alignment in the wild, for wildly successful
languages and formatters:
* [PEP-8](https://peps.python.org/pep-0008/#indentation)
*
[black](https://black.readthedocs.io/en/stable/the_black_code_style/current_style.html#how-black-wraps-lines)
* [prettier](https://prettier.io/docs/en/)
The above examples are useful mainly to show that hanging-indent
_generally_ is no impediment to efficient code reading and on the whole
is an uncontroversial choice as befits the standard library.
why ?
- We already have an emit that does the same thing
- The name asm itself is a bit confusing, you might think it's an alias
for asm.js or something else.
- The asm keyword is used differently on different compiler targets (it
makes it inexpressive).
- Does anyone (other than some compiler libraries) use asm instead of
emit ? If yes, it's a bit strange to use asm somewhere and emit
somewhere. By making the asm keyword for js target deprecated, there
would be even less use of the asm keyword for js target, reducing the
amount of confusion.
- New users might accidentally use a non-universal approach via the asm
keyword instead of emit, and then when they learn about asm, try to
figure out what the differences are.
see https://forum.nim-lang.org/t/10821
---------
Co-authored-by: Andreas Rumpf <rumpf_a@web.de>
fixes#23002, fixes#22841, refs comments in #23097
When an identifier is ambiguous in scope (i.e. multiple imports contain
symbols with the same name), attempt resolving it through type inference
(by creating a symchoice). To do this efficiently, `qualifiedLookUp` had
to be broken up so that `semExpr` can access the ambiguous candidates
directly (now obtained directly via `lookUpCandidates`).
This fixes the linked issues, but an example like:
```nim
let on = 123
{.warning[ProveInit]: on.}
```
will still fail, since `on` is unambiguously the local `let` symbol here
(this is also true for `proc on` but `proc` symbols generate symchoices
anyway).
Type symbols are not considered to not confuse the type inference. This
includes the change in sigmatch, up to this point symchoices with
nonoverloadable symbols could be created, they just wouldn't be
considered during disambiguation. Now every proper symbol except types
are considered in disambiguation, so the correct symbols must be picked
during the creation of the symchoice node. I remember there being a
violating case of this in the compiler, but this was very likely fixed
by excluding type symbols as CI seems to have found no issues.
The pure enum ambiguity test was disabled because ambiguous pure enums
now behave like overloadable enums with this behavior, so we get a
longer error message for `echo amb` like `type mismatch: got <MyEnum |
OtherEnum> but expected T`
Filling in some more logic in `typeRel` that I came across when poking
the compiler in another PR. Some of the cases where `typeRel` returns an
"incorrect" result are actually common, but `sumGeneric` ends up
breaking the tie correctly. There isn't anything wrong with that
necessarily, but I assume that it's preferred these functions behave
just as well in isolation as they do when integrated.
I will be following up this description with specific examples.
