Lets single threaded applications benefit from tracking foreign cells as
well.
After this, `SmallChunk` technically doesn't need to act as a linked
list anymore I think, gotta investigate that more though.
The likelihood of overflowing `chunk.free` also rises, so to work around
that it might make sense to check `foreignCells` instead of adjusting
free space or replace free with a counter for the local capacity.
For Nim compile I can observe a ~10mb reduction, and smaller ones for
other projects.
Still have to look this over some. We'll see. I put sink in this branch
simply because I saw `tyVar` there and for no other reason. In any case
the problem appears to be coming from `liftParamType` as it removes the
`sink` type from the formals.
#23869
Corrects a slicing mistake in the `std/varints` implementation which
caused it to fail when writing large numbers into buffers smaller than
10..13-bytes, now 9-byte buffers are sufficient as the documentation
states.
fixes#19819, fixes#23339
Since #22029 `tyFromExpr` does not match anything in overloading, so
generic bodies can know which call expressions to delay until the type
can be evaluated. However generic type invocations also run overloading
to check for generic constraints even in generic bodies. To prevent them
from failing early from the overload not matching, pretend that
`tyFromExpr` matches. This mirrors the behavior of the compiler in more
basic cases like:
```nim
type
Foo[T: int] = object
x: T
Bar[T] = object
y: Foo[T]
```
Unfortunately this case doesn't respect the constraint (#21181, some
other bugs) but `tyFromExpr` should easily use the same principle when
it does.
Ref: https://github.com/nim-lang/Nim/issues/23788
There was a small leak in the above issue even after fixing the
segfault. The sizes of `free` and `acc` were changed to 32bit because
adding the `foreignCells` field will drastically increase the memory
usage for programs that hold onto memory for a long time if they stay as
64bit.
fixes#23853
Since #22610 generics turns the `Name` in the `GT.Name` expression in
the test code into a sym choice. The problem is when the compiler tries
to instantiate `GT.Name` it also instantiates the sym choice symbols.
`Name` has type `template (E: type ExtensionField)` which contains the
unresolved generic type `ExtensionField`, which the compiler mistakes as
an uninstantiated node, when it's just part of the type of the template.
The compilation of the node itself and hence overloading will handle the
instantiation of the proc, so we avoid instantiating it in `semtypinst`,
similar to how the first nodes of call nodes aren't instantiated.
Honestly, to me the entire design of a (highly!) restricted set of
`FormatLiterals` characters seems antithetical to the very idea of a
format string template. Fixing that is a much larger change, though.
So, this PR just adds `'.'` so that the standard (both input & output!)
notation for decimal numbers in Nim can be used for the seconds part of
a time format in `lib/pure/times.format(.., f)`. It should only make
legal what was illegal and should be harmless since `'.'` is not used in
any special way otherwise.
fixes#23813, partially reverts #23392
Before #23392, if a `gensym` symbol was defined before a proc with the
same name in a template even with an `inject` annotation, the proc would
be `gensym`. After #23392 the proc was instead changed to be `inject` as
long as no `gensym` annotation was given. Now, to keep compatibility
with the old behavior, the behavior is changed back to infer the proc as
`gensym` when no `inject` annotation is given, however an explicit
`inject` annotation will still inject the proc. This is also documented
in the manual as the old behavior was undocumented and the new behavior
is slightly different.
code such as:
```Nim
import std/httpclient # nim c --hint:performance:on
echo newHttpClient(proxy=nil,
headers=newHttpHeaders({"Accept": "*/*"})).getContent("x")
```
(Fix was suggested by @ringabout in a private channel.)
Seems useful since `httpclient` is so basic/probably pervasive with many
hundreds of `import`s across the NimbleVerse.
fixes#3011
In https://github.com/nim-lang/Nim/pull/23532, meta fields that defined
in the object are handled.
In this PR, RefObjectTy is handled as well:
```nim
type
Type = ref object
context: ref object
```
Ref alias won't trigger mata fields checking so there won't have
cascaded errors on `TypeBase`.
```nim
type
TypeBase = object
context: ref object
Type = ref TypeBase
context: ref object
```
The most specific version of `gcd(int,int)` in `std/math` uses bitwise
comparisons from C compilers, which can't be borrowed on the js platform
in the web browser. Conditional compilation here should fix the issue
for this and downstream libraries such as `std/rationals` when compiling
to browser js as the backend.
---------
Co-authored-by: Andreas Rumpf <rumpf_a@web.de>
I've been working on making some basic cuda examples work, both with
cuda (nvcc) and with AMD HIP (hipcc) https://github.com/monofuel/hippo
- hipcc is just a drop-in replacement for clang and works out of the box
with clang settings in Nim. hipcc is capable of compiling for AMD ROCm
or to CUDA, depending on how HIP_PLATFORM is set.
- nvcc is a little quirky. we can use `-x cu` to tell it to handle nim's
`.cpp` files as if they were `.cu` files. nvcc expects all backend
compiler flags to be wrapped with a special `-Xcompiler=""` flag when
compiling and also when linking.
I manually tested on a linux desktop with amd and a laptop with nvidia.
makes new hash the default, with an opt-out (& js-no-big-int) define.
Also update changelog (& fix one typo).
Only really expect the chronos hash-order sensitive test to fail until
they merge that PR and tag a new release.
added small note regarding style insensitivity for parsing enums. the
casing of the first letter is still taken into account for this
function. was confused a little at first because when I read "style
insensitive manner" I thought it meant casing as well and ran into a
couple of `ValueError`'s because of it.
This pr redefines the relation between lambda lifting and closureiter
transformation.
Key takeaways:
- Lambdalifting now has less distinction between closureiters and
regular closures. Namely instead of lifting _all_ closureiter variables,
it lifts only those variables it would also lift for simple closure,
i.e. those not owned by the closure.
- It is now closureiter transformation's responsibility to lift all the
locals that need lifting and are not lifted by lambdalifting. So now we
lift only those locals that appear in more than one state. The rest
remains on stack, yay!
- Closureiter transformation always relies on the closure env param
created by lambdalifting. Special care taken to make lambdalifting
create it even in cases when it's "too early" to lift.
- Environments created by lambdalifting will contain `:state` only for
closureiters, whereas previously any closure env contained it.
IMO this is a more reasonable approach as it simplifies not only
lambdalifting, but transf too (e.g. freshVarsForClosureIters is now gone
for good).
I tried to organize the changes logically by commits, so it might be
easier to review this on per commit basis.
Some ugliness:
- Adding lifting to closureiters transformation I had to repeat this
matching of `return result = value` node. I tried to understand why it
is needed, but that was just another rabbit hole, so I left it for
another time. @Araq your input is welcome.
- In the last commit I've reused currently undocumented `liftLocals`
pragma for symbols so that closureiter transformation will forcefully
lift those even if they don't require lifting otherwise. This is needed
for [yasync](https://github.com/yglukhov/yasync) or else it will be very
sad.
Overall I'm quite happy with the results, I'm seeing some noticeable
code size reductions in my projects. Heavy closureiter/async users,
please give it a go.
(for `string` or any other key type). Independence is nice to ever
change orders. So, change it to just `len` & a `doAssert` like the other
test in the same file.
