Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Produce static/const initializations for variables when possible #12216

Closed
exelotl opened this issue Sep 19, 2019 · 7 comments
Closed

Produce static/const initializations for variables when possible #12216

exelotl opened this issue Sep 19, 2019 · 7 comments
Labels

Comments

@exelotl
Copy link
Contributor

exelotl commented Sep 19, 2019

A top-level initialization in Nim produces C code which first declares the variable, then later assigns a value in the module's Init function.

This is troublesome for embedded programming, because there is no way to make a variable 'const' in the generated C code.

As a result, there is no sensible way to put an object/array in ROM and be able to take a pointer to it using unsafeAddr. See my forum post for why this is important.

Example

The following Nim code:

type Point = object
  x, y: int

let foo* = Point(x:10, y:20)

currently leads to the following C code for our variable 'foo':

// ...

tyObject_Point_R2WHeB5ABtUUFMDs9bCVSuA foo_pp3aIVrMTc9cwqFMF9btpcoQ;

// ...

N_LIB_PRIVATE N_NIMCALL(void, unknown_mymoduleInit000)(void) {
{
	nimZeroMem((void*)(&foo_pp3aIVrMTc9cwqFMF9btpcoQ), sizeof(tyObject_Point_R2WHeB5ABtUUFMDs9bCVSuA));
	foo_pp3aIVrMTc9cwqFMF9btpcoQ.x = ((NI) 10);
	foo_pp3aIVrMTc9cwqFMF9btpcoQ.y = ((NI) 20);
}
}

Proposed Solution

After some further dicussion on IRC, we decided that the best way to handle this would be to make the codegen for let smarter, so that it will produce a fully initialized global or static const in cases where that is possible.

In other words, the previous C code would become:

const tyObject_Point_R2WHeB5ABtUUFMDs9bCVSuA foo_pp3aIVrMTc9cwqFMF9btpcoQ = { ((NI) 10), ((NI) 20) };

I believe this can be done for any top-level let variable as long as the initialization only contains the following:

  • literals / constants
  • addresses of other top-level variables / procedures
  • cast[], sizeof, and any other operations that are acceptable in a static initialization context

The same rules can apply to var too (though I suppose it's less important). The 'const' modifier should be omitted in that case.

However, the codegen should not be applied for any initialization that depends on the result of a procedure or the value of a variable at runtime.

More Examples

Visibility/mutability rules

Nim code:

let a = 1
let b* = 2
var c = 3
var d* = 4

Desired C initializers:

static const NI a_xxx = 1;
const NI b_xxx = 2;
static NI c_xxx = 3;
NI d_xxx = 4;

Using C-style arrays (pointer + size)

Nim code:

type AnimData = object
  loop: bool
  frames: ptr UncheckedArray[int]
  len: int

let walkFrames = [2,3,4]

let walkAnimation = AnimData(
  loop: true,
  frames: cast[ptr UncheckedArray[int]](unsafeAddr walkFrames),
  len: walkFrames.len
)

Desired C initializers:

static const tyArray_xxx walkFrames_xxx = {((NI)2), ((NI)3), ((NI)4)};
static const tyObject_AnimData_xxx walkAnimation_xxx = {
	NIM_TRUE,
	(&walkFrames_xxx),
	((NI) 3)
};

Linked list of procedures

Nim code:

type Action = object
  body: proc () {.nimcall.}
  next: ptr Action

proc fn1() = echo "first"
proc fn2() = echo "second"

let second = Action(body: fn2, next: nil)
let first = Action(body: fn1, next: unsafeAddr second)

Desired C initializers:

static const tyObject_Action_xxx second_xxx = { fn2_xxx, NIM_NIL };
static const tyObject_Action_xxx first_xxx = { fn1_xxx, (&second_xxx) };

Other considerations

  • This change may mess with some people's assumptions about the behavior of unsafeAddr. Currently let variables are still mutable under the hood, but with these changes that would not always be the case.
  • I suspect this wouldn't play nicely with every imported C variable. (See the EACCES example in the manual).
  • It might be worth introducing some pragma to control which kind of codegen is used, in case you (a) really want to guarantee static const (b) really want to avoid it
@Araq Araq added the RFC label Sep 19, 2019
@krux02
Copy link
Contributor

krux02 commented Sep 19, 2019

Well, I looked at the generated code, when I change the let into a const. And this is what the code generator spits out:

NIM_CONST tyObject_Point__r9bGBjWagGEgqMgn2hQfycQ TM__X9bGGOoNyKehOTF7f2xys3g_2 = {((NI) 10), ((NI) 20)};

For some reason the name foo got lost in translation, I don't know yet why that is the case. But this is already almost what you want. The only problem is, that the nim compiler doesn't let you take the address of a const. Well, I tested what happens when I remove the check, and it already works.

I tested what happens when I take the address from the same const from different modules, and it is a different address each time. Constants are instantiated once every time they are used. For small constants, such as PI this is probably not a problem at all, but considering your use case where you want to put your resources into constants, it might become a problems.

So what I am proposing is, that instead of making let work like const, I suggest to make unsafeAddr work on const values, because I think const is much closer to what you want than let. On top, it won't break anybody's assumptions about let.

I don't think it would cover the linked list use case with the pointers. Please elaborate if you really think that one is important.

Then since you want to use constant sections in your binary for resources, there is another idea in my head that I never got time to implement. And that is a feature I know from BlitzBasic, it is explicit resource bundling. Here is an example of that feature in the Monkey2 language. The big advantage here is, the resource data is not generated into C source code. Instead the linker is instructed to bundle the binary data into a section of the binary file. I don't think that you would want to generate hundreds of megabytes of data into C literals and then compile this with a C compiler, when you can just link the data.

I think though that the constant duplication should be removed though. Getting the addres from the same symbol but within different modules should not generated different addresses. This doesn't seem to be a hard problem though. I think we can then also revert to use the original symbol name. People using GDB will be grateful.

@exelotl
Copy link
Contributor Author

exelotl commented Sep 20, 2019

I think allowing unsafeAddr for consts would be fine too, as long as it doesn't create opportunities for bugs (e.g. how would consts containing pointers to other consts work at compile time?). I mentioned something similar in my forum post, but I think Araq is more in favour of solutions around let, I guess for semantic reasons?

For my own use case, I don't particularly need a linked list of pointers to functions (though I wouldn't rule out such a structure being useful), but I do need all the individual parts of that example for different reasons.

  1. I need constants that hold pointers to functions, to implement dispatch tables for scenes and entities in my game.
  2. As far as I'm aware, I need constants with pointers to other constants, in order to have a type like AnimData in my examples (i.e. pointer to array + length of array)

As a separate thing, binary resources sounds like a nice idea too! For my project, the devkitARM makefiles already allow that (I put files in a 'data' folder and it generates .o files). It's handy for raw image data and other stuff that doesn't come from Nim, but somehow their 'bin2o' tool is slower than just using staticRead, so I switched to that instead where possible.

@mratsim
Copy link
Collaborator

mratsim commented Sep 20, 2019

I also prefer adding an {.addressable.} pragma to const.

However I don't see how you can have const pointers even to other consts.

I have 2 use-case for addressable consts:

const
  ExpBits* = 10'i32
  ExpBitsMask* = 1'i32 shl ExpBits

  MantissaBits* = 23'i32
  MantissaBitsMask* = 1'i32 shl 23

  ln2 = ln(2'f32)

  ExpMax* = 88'i32
  ExpMin* = -88'i32
  ExpA* = ExpBitsMask.float32 / ln2
  ExpB* = ln2 / ExpBitsMask.float32

func initExpLUT(): array[ExpBitsMask, int32] =
  for i, val in result.mpairs:
    let y = pow(2'f32, i.float32 / ExpBitsMask.float32)
    val = cast[int32](y) and (MantissaBitsMask - 1)

# We need ExpLUT in the BSS so that we can take it's address so it can't be const
let ExpLUT* = initExpLUT()
type
  Fp*[P: static BigInt] = object
    ## P is a prime number
    ## All operations on a field are modulo P
    value: type(P)

  Montgomery*[M: static BigInt] = object
    ## All operations in the Montgomery domain
    ## are modulo M. M **must** be odd
    value: type(M)

@cooldome
Copy link
Member

IMO, it is easier to implement addressable const but it is a step back. It would help with this use case but others will complain as it is not always a win. While using const for let would some be something everyone will benefit.

@mratsim
Copy link
Collaborator

mratsim commented Sep 22, 2019

Those are not mutually exclusive though.

On another note, with let foo {.compileTime.} being visible at runtime (which I don't really agree with but that's another issue), is the data inlined at use site or is it stored in the binary and then a pointer is used, for example for an array.

@arnetheduck
Copy link
Contributor

arnetheduck commented Sep 24, 2019

fwiw, nlvm already does this for let globals, and it generates code similar to const <type> = <init> for all let. It doesn't use the equivalent of static because afair that may cause two symbols with the same content within a compilation unit share address - this is fine for const, but not for let (ie that's one significant difference between then).

there are other differences: const-style stuff is generally expected to be inlined at compile time while let should be linked - I don't think I've seen this used practically in nim, but basically if you compile a static library and change a let value, the value from the static library should be used. this is similar to putting a constant in the header vs creating a symbol for it and linking it in, in C.

thus for let *, it seems that it should use const in the declaring module and extern const; in others to ensure it's the same symbol across the compilation (nlvm compiles everything as a single module so it doesn't have this issue), depending on desired address nuances.

@arnetheduck
Copy link
Contributor

http://llvm.org/docs/LangRef.html#linkage-types has a good reference on the various linking options and what they mean / imply. here's the corresponding llvm IR that gets generated for c - can play around with different options easily: https://gcc.godbolt.org/z/1VS0vJ

@Araq Araq closed this as completed in 0e7338d Dec 5, 2019
foldl added a commit to foldl/Nim that referenced this issue Dec 6, 2019
This is the missing bit of PR nim-lang#12799, and fixes issue nim-lang#12216.
Araq added a commit that referenced this issue Dec 10, 2019
Araq added a commit that referenced this issue Dec 10, 2019
netbsd-srcmastr pushed a commit to NetBSD/pkgsrc that referenced this issue Apr 5, 2020
Changelog:
# v1.2.0 - 2020-04-02

## Standard library additions and changes

- Added overloaded `strformat.fmt` macro that use specified characters as
  delimiter instead of '{' and '}'.
- Added new procs in `tables.nim`: `OrderedTable.pop`, `CountTable.del`,
  `CountTable.pop`, `Table.pop`.
- Added `strtabs.clear` overload that reuses the existing mode.
- Added `browsers.osOpen` const alias for the operating system specific *"open"* command.
- Added `sugar.dup` for turning in-place algorithms like `sort` and `shuffle`
  into operations that work on a copy of the data and return the mutated copy,
  like the existing `sorted` does.
- Added `sugar.collect` that does comprehension for seq/set/table collections.
- Added `sugar.capture` for capturing some local loop variables when creating a
  closure. This is an enhanced version of `closureScope`.
- Added `typetraits.tupleLen` to get number of elements of a tuple/type tuple,
  and `typetraits.get` to get the ith element of a type tuple.
- Added `typetraits.genericParams` to return a tuple of generic params from a
  generic instantiation.
- `options` now treats `proc` like other pointer types, meaning `nil` proc variables
  are converted to `None`.
- Added `os.normalizePathEnd` for additional path sanitization.
- Added `times.fromUnixFloat,toUnixFloat`, sub-second resolution versions of
  `fromUnix`,`toUnixFloat`.
- Added `wrapnils` module for chains of field-access and indexing where the LHS
  can be nil. This simplifies code by reducing need for if-else branches around
  intermediate maybe nil values. E.g. `echo ?.n.typ.kind`.
- Added `minIndex`, `maxIndex` and `unzip` to the `sequtils` module.
- Added `os.isRelativeTo` to tell whether a path is relative to another.
- Added `resetOutputFormatters` to `unittest`.
- Added `expectIdent` to the `macros` module.
- Added `os.isValidFilename` that returns `true` if `filename` argument is valid
  for cross-platform use.
- Added `times.isLeapDay`
- `base64` adds URL-Safe Base64, implements RFC-4648 Section-7.
- Added a new module, `std / compilesettings` for querying the compiler about
  diverse configuration settings.
- Added `net.getPeerCertificates` and `asyncnet.getPeerCertificates` for
  retrieving the verified certificate chain of the peer we are connected to
  through an SSL-wrapped `Socket`/`AsyncSocket`.
- Added `browsers.openDefaultBrowser` without URL, implements IETF RFC-6694 Section-3.
- Added `jsconsole.trace`, `jsconsole.table`, `jsconsole.exception` for JavaScript target.
- Added `distinctBase` overload for values: `assert 12.MyInt.distinctBase == 12`
- Added new module `std/stackframes`, in particular `setFrameMsg`, which enables
  custom runtime annotation of stackframes, see #13351 for examples.
  Turn on/off via `--stackTraceMsgs:on/off`.
- Added `sequtils.countIt`, allowing for counting items using a predicate.
- Added a `with` macro for easy function chaining that's available everywhere,
  there is no need to concern your APIs with returning the first argument
  to enable "chaining", instead use the dedicated macro `with` that
  was designed for it. For example:

```nim
type
  Foo = object
    col, pos: string

proc setColor(f: var Foo; r, g, b: int) = f.col = $(r, g, b)
proc setPosition(f: var Foo; x, y: float) = f.pos = $(x, y)

var f: Foo
with(f, setColor(2, 3, 4), setPosition(0.0, 1.0))
echo f
```

- `macros.newLit` now works for ref object types.
- `macro pragmas` can now be used in type sections.
- 5 new pragmas were added to Nim in order to make the upcoming tooling more
  convenient to use. Nim compiler checks these pragmas for syntax but otherwise
  ignores them. The pragmas are `requires`, `ensures`, `assume`, `assert`, `invariant`.
- `system.writeFile` has been overloaded to also support `openarray[byte]`.
- `asyncdispatch.drain` now properly takes into account `selector.hasPendingOperations`
  and only returns once all pending async operations are guaranteed to have completed.
- `sequtils.zip` now returns a sequence of anonymous tuples i.e. those tuples
  now do not have fields named "a" and "b".
- `distinctBase` has been moved from `sugar` to `typetraits` and now it is
  implemented as compiler type trait instead of macro. `distinctBase` in sugar
  module is now deprecated.
- `CountTable.mget` has been removed from `tables.nim`. It didn't work, and it
  was an oversight to be included in v1.0.
- `tables.merge(CountTable, CountTable): CountTable` has been removed.
  It didn't work well together with the existing inplace version of the same proc
  (`tables.merge(var CountTable, CountTable)`).
  It was an oversight to be included in v1.0.
- `asyncdispatch.drain` now consistently uses the passed timeout value for all
  iterations of the event loop, and not just the first iteration.
  This is more consistent with the other asyncdispatch APIs, and allows
  `asyncdispatch.drain` to be more efficient.
- `base64.encode` and `base64.decode` were made faster by about 50%.
- `htmlgen` adds [MathML](https://wikipedia.org/wiki/MathML) support
  (ISO 40314).
- `macros.eqIdent` is now invariant to export markers and backtick quotes.
- `htmlgen.html` allows `lang` in the `<html>` tag and common valid attributes.
- `macros.basename` and `basename=` got support for `PragmaExpr`,
  so that an expression like `MyEnum {.pure.}` is handled correctly.
- `httpclient.maxredirects` changed from `int` to `Natural`, because negative values
  serve no purpose whatsoever.
- `httpclient.newHttpClient` and `httpclient.newAsyncHttpClient` added `headers`
  argument to set initial HTTP Headers, instead of a hardcoded empty `newHttpHeader()`.
- `parseutils.parseUntil` has now a different behaviour if the `until` parameter is
  empty. This was required for intuitive behaviour of the strscans module
  (see bug #13605).
- `strutils.formatFloat` with `precision = 0` has the same behavior in all
  backends, and it is compatible with Python's behavior,
  e.g. `formatFloat(3.14159, precision = 0)` is now `3`, not `3.`.
- `times.parse` now only uses input to compute its result, and not `now`:
  `parse("2020", "YYYY", utc())` is now `2020-01-01T00:00:00Z` instead of
  `2020-03-02T00:00:00Z` if run on 03-02; it also doesn't crash anymore when
  used on 29th, 30th, 31st of each month.
- `httpcore.==(string, HttpCode)` is now deprecated due to lack of practical
  usage. The `$` operator can be used to obtain the string form of `HttpCode`
  for comparison if desired.
- `std/oswalkdir` was buggy, it's now deprecated and reuses `std/os` procs.
- `net.newContext` now performs SSL Certificate checking on Linux and OSX.
  Define `nimDisableCertificateValidation` to disable it globally.
- `os.walkDir` and `os.walkDirRec` now have new flag, `checkDir` (default: false).
  If it is set to true, it will throw if input dir is invalid instead of a noop
  (which is the default behaviour, as it was before this change),
  `os.walkDirRec` only throws if top-level dir is invalid, but ignores errors for
  subdirs, otherwise it would be impossible to resume iteration.
- The `FD` variant of `selector.unregister` for `ioselector_epoll` and
  `ioselector_select` now properly handle the `Event.User` select event type.
- `joinPath` path normalization when `/` is the first argument works correctly:
  `assert "/" / "/a" == "/a"`. Fixed the edge case: `assert "" / "" == ""`.
- `xmltree` now adds indentation consistently to child nodes for any number
  of children nodes.
- `os.splitPath()` behavior synchronized with `os.splitFile()` to return "/"
  as the dir component of `/root_sub_dir` instead of the empty string.
- The deprecated `lc` macro has been removed from `sugar`. It is now replaced with the
  more powerful `collect` macro.
- `os.relativePath("foo", "foo")` is now `"."`, not `""`, as `""` means invalid
  path and shouldn't be conflated with `"."`; use `-d:nimOldRelativePathBehavior`
  to restore the old behavior.
- `os.joinPath(a, b)` now honors trailing slashes in `b` (or `a` if `b` = "").
- `base64.encode` no longer supports `lineLen` and `newLine`.
  Use `base64.encodeMime` instead.


## Language changes

- An `align` pragma can now be used for variables and object fields, similar
  to the `alignas` declaration modifier in C/C++.
- `=sink` type bound operator is now optional. Compiler can now use combination
  of `=destroy` and `copyMem` to move objects efficiently.
- Unsigned integer operators have been fixed to allow promotion of the first operand.
- Conversions to unsigned integers are unchecked at runtime, imitating earlier Nim
  versions. The documentation was improved to acknowledge this special case.
  See nim-lang/RFCs#175 for more details.
- New syntax for lvalue references: `var b {.byaddr.} = expr` enabled by
  `import std/decls`.
- `var a {.foo.}: MyType = expr` now lowers to `foo(a, MyType, expr)` for
  non-builtin pragmas, enabling things like lvalue references (see `decls.byaddr`).


## Compiler changes

- Generated JS code uses spaces, instead of mixed spaces and tabs.
- The Nim compiler now supports the ``--asm`` command option for easier
  inspection of the produced assembler code.
- The Nim compiler now supports a new pragma called ``.localPassc`` to
  pass specific compiler options to the C(++) backend for the C(++) file
  that was produced from the current Nim module.
- The compiler now inferes "sink parameters". To disable this for a specific routine,
  annotate it with `.nosinks`. To disable it for a section of code, use
  `{.push sinkInference: off.}`...`{.pop.}`.
- The compiler now supports a new switch `--panics:on` that turns runtime
  errors like `IndexError` or `OverflowError` into fatal errors that **cannot**
  be caught via Nim's `try` statement. `--panics:on` can improve the
  runtime efficiency and code size of your program significantly.
- The compiler now warns about inheriting directly from `system.Exception` as
  this is **very bad** style. You should inherit from `ValueError`, `IOError`,
  `OSError` or from a different specific exception type that inherits from
  `CatchableError` and cannot be confused with a `Defect`.
- The error reporting for Nim's effect system has been improved.
- Implicit conversions for `const` behave correctly now, meaning that code like
  `const SOMECONST = 0.int; procThatTakesInt32(SOMECONST)` will be illegal now.
  Simply write `const SOMECONST = 0` instead.
- The `{.dynlib.}` pragma is now required for exporting symbols when making
  shared objects on POSIX and macOS, which make it consistent with the behavior
  on Windows.
- The compiler is now more strict about type conversions concerning proc
  types: Type conversions cannot be used to hide `.raise` effects or side
  effects, instead a `cast` must be used. With the flag `--useVersion:1.0` the
  old behaviour is emulated.
- The Nim compiler now implements a faster way to detect overflows based
  on GCC's `__builtin_sadd_overflow` family of functions. (Clang also
  supports these). Some versions of GCC lack this feature and unfortunately
  we cannot detect this case reliably. So if you get compilation errors like
  "undefined reference to `__builtin_saddll_overflow`" compile your programs
  with `-d:nimEmulateOverflowChecks`.


## Tool changes

- Nimpretty doesn't accept negative indentation argument anymore, because it was
  breaking files.


## Bugfixes

- Fixed "`nimgrep --nocolor` is ignored on posix; should be instead: `--nimgrep --color=[auto]|true|false`"
  ([#7591](nim-lang/Nim#7591))
- Fixed "Runtime index on const array (of converted obj) causes C-compiler error"
  ([#10514](nim-lang/Nim#10514))
- Fixed "windows x86 with vcc compile error with "asmNoStackFrame""
  ([#12298](nim-lang/Nim#12298))
- Fixed "[TODO] regression: Error: Locks requires --threads:on option"
  ([#12330](nim-lang/Nim#12330))
- Fixed "Add --cc option to --help or --fullhelp output"
  ([#12010](nim-lang/Nim#12010))
- Fixed "questionable `csize` definition in `system.nim`"
  ([#12187](nim-lang/Nim#12187))
- Fixed "os.getAppFilename() returns incorrect results on OpenBSD"
  ([#12389](nim-lang/Nim#12389))
- Fixed "HashSet[uint64] slow insertion depending on values"
  ([#11764](nim-lang/Nim#11764))
- Fixed "Differences between compiling 'classic call syntax' vs 'method call syntax' ."
  ([#12453](nim-lang/Nim#12453))
- Fixed "c -d:nodejs --> SIGSEGV: Illegal storage access"
  ([#12502](nim-lang/Nim#12502))
- Fixed "Closure iterator crashes on --newruntime due to "dangling references""
  ([#12443](nim-lang/Nim#12443))
- Fixed "No `=destroy` for elements of closure environments other than for latest devel --gc:destructors"
  ([#12577](nim-lang/Nim#12577))
- Fixed "strutils:formatBiggestFloat() gives different results in JS mode"
  ([#8242](nim-lang/Nim#8242))
- Fixed "Regression (devel): the new `csize_t` definition isn't consistently used, nor tested thoroughly..."
  ([#12597](nim-lang/Nim#12597))
- Fixed "tables.take() is defined only for `Table` and missed for other table containers"
  ([#12519](nim-lang/Nim#12519))
- Fixed "`pthread_key_t` errors on OpenBSD"
  ([#12135](nim-lang/Nim#12135))
- Fixed "newruntime: simple seq pop at ct results in compile error"
  ([#12644](nim-lang/Nim#12644))
- Fixed "[Windows] finish.exe C:\Users\<USERNAME>\.nimble\bin is not in your PATH environment variable."
  ([#12319](nim-lang/Nim#12319))
- Fixed "Error with strformat + asyncdispatch + const"
  ([#12612](nim-lang/Nim#12612))
- Fixed "MultipartData needs $"
  ([#11863](nim-lang/Nim#11863))
- Fixed "Nim stdlib style issues with --styleCheck:error"
  ([#12687](nim-lang/Nim#12687))
- Fixed "new $nimbleDir path substitution yields unexpected search paths"
  ([#12767](nim-lang/Nim#12767))
- Fixed "Regression: inlined procs now get multiple rounds of destructor injection"
  ([#12766](nim-lang/Nim#12766))
- Fixed "newruntime: compiler generates defective code"
  ([#12669](nim-lang/Nim#12669))
- Fixed "broken windows modules path handling because of 'os.relativePath' breaking changes"
  ([#12734](nim-lang/Nim#12734))
- Fixed "for loop tuple syntax not rendered correctly"
  ([#12740](nim-lang/Nim#12740))
- Fixed "Crash when trying to use `type.name[0]`"
  ([#12804](nim-lang/Nim#12804))
- Fixed "Enums should be considered Trivial types in Atomics"
  ([#12812](nim-lang/Nim#12812))
- Fixed "Produce static/const initializations for variables when possible"
  ([#12216](nim-lang/Nim#12216))
- Fixed "Assigning descriminator field leads to internal assert with --gc:destructors"
  ([#12821](nim-lang/Nim#12821))
- Fixed "nimsuggest `use` command does not return all instances of symbol"
  ([#12832](nim-lang/Nim#12832))
- Fixed "@[] is a problem for --gc:destructors"
  ([#12820](nim-lang/Nim#12820))
- Fixed "Codegen ICE in allPathsAsgnResult"
  ([#12827](nim-lang/Nim#12827))
- Fixed "seq[Object with ref and destructor type] doesn't work in old runtime"
  ([#12882](nim-lang/Nim#12882))
- Fixed "Destructor not invoked because it is instantiated too late, old runtime"
  ([#12883](nim-lang/Nim#12883))
- Fixed "The collect macro does not handle if/case correctly"
  ([#12874](nim-lang/Nim#12874))
- Fixed "allow typed/untyped params in magic procs (even if not in stdlib)"
  ([#12911](nim-lang/Nim#12911))
- Fixed "ARC/newruntime memory corruption"
  ([#12899](nim-lang/Nim#12899))
- Fixed "tasyncclosestall.nim still flaky test: Address already in use"
  ([#12919](nim-lang/Nim#12919))
- Fixed "newruntime and computed goto: variables inside the loop are in generated code uninitialised"
  ([#12785](nim-lang/Nim#12785))
- Fixed "osx: dsymutil needs to be called for debug builds to keep debug info"
  ([#12735](nim-lang/Nim#12735))
- Fixed "codegen ICE with ref objects, gc:destructors"
  ([#12826](nim-lang/Nim#12826))
- Fixed "mutable iterator cannot yield named tuples"
  ([#12945](nim-lang/Nim#12945))
- Fixed "parsecfg stores "\r\n" line breaks just as "\n""
  ([#12970](nim-lang/Nim#12970))
- Fixed "db_postgres.getValue issues warning when no rows found"
  ([#12973](nim-lang/Nim#12973))
- Fixed "ARC: Unpacking tuple with seq causes segfault"
  ([#12989](nim-lang/Nim#12989))
- Fixed "ARC/newruntime: strutils.join on seq with only empty strings causes segfault"
  ([#12965](nim-lang/Nim#12965))
- Fixed "regression (1.0.4): `{.push exportc.}` wrongly affects generic instantiations, causing codegen errors"
  ([#12985](nim-lang/Nim#12985))
- Fixed "cdt, crash with --gc:arc, works fine with default gc"
  ([#12978](nim-lang/Nim#12978))
- Fixed "ARC: No indexError thrown on out-of-bound seq access, SIGSEGV instead"
  ([#12961](nim-lang/Nim#12961))
- Fixed "ARC/async: Returning in a try-block results in wrong codegen"
  ([#12956](nim-lang/Nim#12956))
- Fixed "asm keyword is generating wrong output C code when --cc:tcc"
  ([#12988](nim-lang/Nim#12988))
- Fixed "Destructor not invoked"
  ([#13026](nim-lang/Nim#13026))
- Fixed "ARC/newruntime: Adding inherited var ref object to seq with base type causes segfault"
  ([#12964](nim-lang/Nim#12964))
- Fixed "Style check error with JS compiler target"
  ([#13032](nim-lang/Nim#13032))
- Fixed "regression(1.0.4): undeclared identifier: 'readLines'; plus another regression and bug"
  ([#13013](nim-lang/Nim#13013))
- Fixed "regression(1.04) `invalid pragma: since` with nim js"
  ([#12996](nim-lang/Nim#12996))
- Fixed "Sink to MemMove optimization in injectdestructors"
  ([#13002](nim-lang/Nim#13002))
- Fixed "--gc:arc: `catch` doesn't work with exception subclassing"
  ([#13072](nim-lang/Nim#13072))
- Fixed "nim c --gc:arc --exceptions:{setjmp,goto} incorrectly handles raise; `nim cpp --gc:arc` is ok"
  ([#13070](nim-lang/Nim#13070))
- Fixed "typetraits feature request - get subtype of a generic type"
  ([#6454](nim-lang/Nim#6454))
- Fixed "CountTable inconsistencies between keys() and len() after setting value to 0"
  ([#12813](nim-lang/Nim#12813))
- Fixed "{.align.} pragma is not applied if there is a generic field"
  ([#13122](nim-lang/Nim#13122))
- Fixed "ARC, finalizer, allow rebinding the same function multiple times"
  ([#13112](nim-lang/Nim#13112))
- Fixed "`nim doc` treats `export localSymbol` incorrectly"
  ([#13100](nim-lang/Nim#13100))
- Fixed "--gc:arc SIGSEGV (double free?)"
  ([#13119](nim-lang/Nim#13119))
- Fixed "codegen bug with arc"
  ([#13105](nim-lang/Nim#13105))
- Fixed "symbols not defined in the grammar"
  ([#10665](nim-lang/Nim#10665))
- Fixed "[JS] Move is not defined"
  ([#9674](nim-lang/Nim#9674))
- Fixed "[TODO] pathutils.`/` can return invalid AbsoluteFile"
  ([#13121](nim-lang/Nim#13121))
- Fixed "regression(1.04) `nim doc main.nim` generates broken html (no css)"
  ([#12998](nim-lang/Nim#12998))
- Fixed "Wrong supportsCopyMem on string in type section"
  ([#13095](nim-lang/Nim#13095))
- Fixed "Arc, finalizer, out of memory"
  ([#13157](nim-lang/Nim#13157))
- Fixed "`--genscript` messes up nimcache and future nim invocations"
  ([#13144](nim-lang/Nim#13144))
- Fixed "--gc:arc with --exceptions:goto for "nim c" generate invalid c code"
  ([#13186](nim-lang/Nim#13186))
- Fixed "[regression] duplicate member `_i1` codegen bug"
  ([#13195](nim-lang/Nim#13195))
- Fixed "RTree investigations with valgrind for --gc:arc"
  ([#13110](nim-lang/Nim#13110))
- Fixed "relativePath("foo", ".") returns wrong path"
  ([#13211](nim-lang/Nim#13211))
- Fixed "asyncftpclient - problem with welcome.msg"
  ([#4684](nim-lang/Nim#4684))
- Fixed "Unclear error message, lowest priority"
  ([#13256](nim-lang/Nim#13256))
- Fixed "Channel messages are corrupted"
  ([#13219](nim-lang/Nim#13219))
- Fixed "Codegen bug with exportc and case objects"
  ([#13281](nim-lang/Nim#13281))
- Fixed "[bugfix] fix #11590: c compiler warnings silently ignored, giving undefined behavior"
  ([#11591](nim-lang/Nim#11591))
- Fixed "[CI] tnetdial flaky test"
  ([#13132](nim-lang/Nim#13132))
- Fixed "Cross-Compiling with -d:mingw fails to locate compiler under OSX"
  ([#10717](nim-lang/Nim#10717))
- Fixed "`nim doc --project` broken with imports below main project file or duplicate names"
  ([#13150](nim-lang/Nim#13150))
- Fixed "regression: isNamedTuple(MyGenericTuple[int]) is false, should be true"
  ([#13349](nim-lang/Nim#13349))
- Fixed "--gc:arc codegen bug copying objects bound to C structs with missing C struct fields"
  ([#13269](nim-lang/Nim#13269))
- Fixed "write requires conversion to string"
  ([#13182](nim-lang/Nim#13182))
- Fixed "Some remarks to stdlib documentation"
  ([#13352](nim-lang/Nim#13352))
- Fixed "a `check` in unittest generated by template doesn't show actual value"
  ([#6736](nim-lang/Nim#6736))
- Fixed "Implicit return with case expression fails with 'var' return."
  ([#3339](nim-lang/Nim#3339))
- Fixed "Segfault with closure on arc"
  ([#13314](nim-lang/Nim#13314))
- Fixed "[Macro] Crash on malformed case statement with multiple else"
  ([#13255](nim-lang/Nim#13255))
- Fixed "regression: `echo 'discard' | nim c -r -` generates a file '-' ; `-` should be treated specially"
  ([#13374](nim-lang/Nim#13374))
- Fixed "on OSX, debugging (w gdb or lldb) a nim program crashes at the 1st call to `execCmdEx`"
  ([#9634](nim-lang/Nim#9634))
- Fixed "Internal error in getTypeDescAux"
  ([#13378](nim-lang/Nim#13378))
- Fixed "gc:arc mode breaks tuple let"
  ([#13368](nim-lang/Nim#13368))
- Fixed "Nim compiler hangs for certain C/C++ compiler errors"
  ([#8648](nim-lang/Nim#8648))
- Fixed "htmlgen does not support `data-*` attributes"
  ([#13444](nim-lang/Nim#13444))
- Fixed "[gc:arc] setLen will cause string not to be null-terminated."
  ([#13457](nim-lang/Nim#13457))
- Fixed "joinPath("", "") is "/" ; should be """
  ([#13455](nim-lang/Nim#13455))
- Fixed "[CI] flaky test on windows: tests/osproc/texitcode.nim"
  ([#13449](nim-lang/Nim#13449))
- Fixed "Casting to float32 on NimVM is broken"
  ([#13479](nim-lang/Nim#13479))
- Fixed "`--hints:off` doesn't work (doesn't override ~/.config/nim.cfg)"
  ([#8312](nim-lang/Nim#8312))
- Fixed "joinPath("", "") is "/" ; should be """
  ([#13455](nim-lang/Nim#13455))
- Fixed "tables.values is broken"
  ([#13496](nim-lang/Nim#13496))
- Fixed "global user config can override project specific config"
  ([#9405](nim-lang/Nim#9405))
- Fixed "Non deterministic macros and id consistency problem"
  ([#12627](nim-lang/Nim#12627))
- Fixed "try expression doesn't work with return on expect branch"
  ([#13490](nim-lang/Nim#13490))
- Fixed "CI will break every 4 years on feb 28: times doesn't handle leap years properly"
  ([#13543](nim-lang/Nim#13543))
- Fixed "[minor] `nimgrep --word` doesn't work with operators (eg misses  `1 +% 2`)"
  ([#13528](nim-lang/Nim#13528))
- Fixed "`as` is usable as infix operator but its existence and precedence are not documented"
  ([#13409](nim-lang/Nim#13409))
- Fixed "JSON unmarshalling drops seq's items"
  ([#13531](nim-lang/Nim#13531))
- Fixed "os.joinPath returns wrong path when head ends '\' or '/' and tail starts '..'."
  ([#13579](nim-lang/Nim#13579))
- Fixed "Block-local types with the same name lead to bad codegen (sighashes regression)"
  ([#5170](nim-lang/Nim#5170))
- Fixed "tuple codegen error"
  ([#12704](nim-lang/Nim#12704))
- Fixed "newHttpHeaders does not accept repeated headers"
  ([#13573](nim-lang/Nim#13573))
- Fixed "regression: --incremental:on fails on simplest example"
  ([#13319](nim-lang/Nim#13319))
- Fixed "strscan can't get value of last element in format"
  ([#13605](nim-lang/Nim#13605))
- Fixed "hashes_examples crashes with "Bus Error" (unaligned access) on sparc64"
  ([#12508](nim-lang/Nim#12508))
- Fixed "gc:arc bug with re-used `seq[T]`"
  ([#13596](nim-lang/Nim#13596))
- Fixed "`raise CatchableError` is broken with --gc:arc  when throwing inside a proc"
  ([#13599](nim-lang/Nim#13599))
- Fixed "cpp --gc:arc --exceptions:goto fails to raise with discard"
  ([#13436](nim-lang/Nim#13436))
- Fixed "terminal doesn't compile with -d:useWinAnsi"
  ([#13607](nim-lang/Nim#13607))
- Fixed "Parsing "sink ptr T" - region needs to be an object type"
  ([#12757](nim-lang/Nim#12757))
- Fixed "gc:arc + threads:on + closures compilation error"
  ([#13519](nim-lang/Nim#13519))
- Fixed "[ARC] segmentation fault"
  ([#13240](nim-lang/Nim#13240))
- Fixed "times.toDateTime buggy on 29th, 30th and 31th of each month"
  ([#13558](nim-lang/Nim#13558))
- Fixed "Deque misbehaves on VM"
  ([#13310](nim-lang/Nim#13310))
- Fixed "Nimscript listFiles should throw exception when path is not found"
  ([#12676](nim-lang/Nim#12676))
- Fixed "koch boot fails if even an empty config.nims is present in ~/.config/nims/ [devel regression]"
  ([#13633](nim-lang/Nim#13633))
- Fixed "nim doc generates lots of false positive LockLevel warnings"
  ([#13218](nim-lang/Nim#13218))
- Fixed "Arrays are passed by copy to iterators, causing crashes, unnecessary allocations and slowdowns"
  ([#12747](nim-lang/Nim#12747))
- Fixed "Range types always uses signed integer as a base type"
  ([#13646](nim-lang/Nim#13646))
- Fixed "Generate c code cannot compile with recent devel version"
  ([#13645](nim-lang/Nim#13645))
- Fixed "[regression] VM: Error: cannot convert -1 to uint64"
  ([#13661](nim-lang/Nim#13661))
- Fixed "Spurious raiseException(Exception) detected"
  ([#13654](nim-lang/Nim#13654))
- Fixed "gc:arc memory leak"
  ([#13659](nim-lang/Nim#13659))
- Fixed "Error: cannot convert -1 to uint (inside tuples)"
  ([#13671](nim-lang/Nim#13671))
- Fixed "strformat issue with --gc:arc"
  ([#13622](nim-lang/Nim#13622))
- Fixed "astToStr doesn't work inside generics"
  ([#13524](nim-lang/Nim#13524))
- Fixed "oswalkdir.walkDirRec wont return folders"
  ([#11458](nim-lang/Nim#11458))
- Fixed "`echo 'echo 1' | nim c -r -`  silently gives wrong results (nimBetterRun not updated for stdin)"
  ([#13412](nim-lang/Nim#13412))
- Fixed "gc:arc destroys the global variable accidentally."
  ([#13691](nim-lang/Nim#13691))
- Fixed "[minor] sigmatch errors should be sorted, for reproducible errors"
  ([#13538](nim-lang/Nim#13538))
- Fixed "Exception when converting csize to clong"
  ([#13698](nim-lang/Nim#13698))
- Fixed "ARC: variables are no copied on the thread spawn causing crashes"
  ([#13708](nim-lang/Nim#13708))
- Fixed "Illegal distinct seq causes compiler crash"
  ([#13720](nim-lang/Nim#13720))
- Fixed "cyclic seq definition crashes the compiler"
  ([#13715](nim-lang/Nim#13715))
- Fixed "Iterator with openArray parameter make the argument evaluated many times"
  ([#13417](nim-lang/Nim#13417))
- Fixed "net/asyncnet: Unable to access peer's certificate chain"
  ([#13299](nim-lang/Nim#13299))
- Fixed "Accidentally "SIGSEGV: Illegal storage access" error after arc optimizations (#13325)"
  ([#13709](nim-lang/Nim#13709))
- Fixed "Base64 Regression"
  ([#13722](nim-lang/Nim#13722))
- Fixed "A regression (?) with --gc:arc and repr"
  ([#13731](nim-lang/Nim#13731))
- Fixed "Internal compiler error when using the new variable pragmas"
  ([#13737](nim-lang/Nim#13737))
- Fixed "bool conversion produces vcc 2019 warning at cpp compilation stage"
  ([#13744](nim-lang/Nim#13744))
- Fixed "Compiler "does not detect" a type recursion error in the wrong code, remaining frozen"
  ([#13763](nim-lang/Nim#13763))
- Fixed "[minor] regression: `Foo[0.0] is Foo[-0.0]` is now false"
  ([#13730](nim-lang/Nim#13730))
- Fixed "`nim doc` - only whitespace on first line causes segfault"
  ([#13631](nim-lang/Nim#13631))
- Fixed "hashset regression"
  ([#13794](nim-lang/Nim#13794))
- Fixed "`os.getApplFreebsd` could return incorrect paths in the case of a long path"
  ([#13806](nim-lang/Nim#13806))
- Fixed "Destructors are not inherited"
  ([#13810](nim-lang/Nim#13810))
- Fixed "io.readLines AssertionError on devel"
  ([#13829](nim-lang/Nim#13829))
- Fixed "exceptions:goto accidentally reset the variable during exception handling"
  ([#13782](nim-lang/Nim#13782))
netbsd-srcmastr pushed a commit to NetBSD/pkgsrc that referenced this issue Apr 6, 2020
Changelog:
# v1.2.0 - 2020-04-02

## Standard library additions and changes

- Added overloaded `strformat.fmt` macro that use specified characters as
  delimiter instead of '{' and '}'.
- Added new procs in `tables.nim`: `OrderedTable.pop`, `CountTable.del`,
  `CountTable.pop`, `Table.pop`.
- Added `strtabs.clear` overload that reuses the existing mode.
- Added `browsers.osOpen` const alias for the operating system specific *"open"* command.
- Added `sugar.dup` for turning in-place algorithms like `sort` and `shuffle`
  into operations that work on a copy of the data and return the mutated copy,
  like the existing `sorted` does.
- Added `sugar.collect` that does comprehension for seq/set/table collections.
- Added `sugar.capture` for capturing some local loop variables when creating a
  closure. This is an enhanced version of `closureScope`.
- Added `typetraits.tupleLen` to get number of elements of a tuple/type tuple,
  and `typetraits.get` to get the ith element of a type tuple.
- Added `typetraits.genericParams` to return a tuple of generic params from a
  generic instantiation.
- `options` now treats `proc` like other pointer types, meaning `nil` proc variables
  are converted to `None`.
- Added `os.normalizePathEnd` for additional path sanitization.
- Added `times.fromUnixFloat,toUnixFloat`, sub-second resolution versions of
  `fromUnix`,`toUnixFloat`.
- Added `wrapnils` module for chains of field-access and indexing where the LHS
  can be nil. This simplifies code by reducing need for if-else branches around
  intermediate maybe nil values. E.g. `echo ?.n.typ.kind`.
- Added `minIndex`, `maxIndex` and `unzip` to the `sequtils` module.
- Added `os.isRelativeTo` to tell whether a path is relative to another.
- Added `resetOutputFormatters` to `unittest`.
- Added `expectIdent` to the `macros` module.
- Added `os.isValidFilename` that returns `true` if `filename` argument is valid
  for cross-platform use.
- Added `times.isLeapDay`
- `base64` adds URL-Safe Base64, implements RFC-4648 Section-7.
- Added a new module, `std / compilesettings` for querying the compiler about
  diverse configuration settings.
- Added `net.getPeerCertificates` and `asyncnet.getPeerCertificates` for
  retrieving the verified certificate chain of the peer we are connected to
  through an SSL-wrapped `Socket`/`AsyncSocket`.
- Added `browsers.openDefaultBrowser` without URL, implements IETF RFC-6694 Section-3.
- Added `jsconsole.trace`, `jsconsole.table`, `jsconsole.exception` for JavaScript target.
- Added `distinctBase` overload for values: `assert 12.MyInt.distinctBase == 12`
- Added new module `std/stackframes`, in particular `setFrameMsg`, which enables
  custom runtime annotation of stackframes, see #13351 for examples.
  Turn on/off via `--stackTraceMsgs:on/off`.
- Added `sequtils.countIt`, allowing for counting items using a predicate.
- Added a `with` macro for easy function chaining that's available everywhere,
  there is no need to concern your APIs with returning the first argument
  to enable "chaining", instead use the dedicated macro `with` that
  was designed for it. For example:

```nim
type
  Foo = object
    col, pos: string

proc setColor(f: var Foo; r, g, b: int) = f.col = $(r, g, b)
proc setPosition(f: var Foo; x, y: float) = f.pos = $(x, y)

var f: Foo
with(f, setColor(2, 3, 4), setPosition(0.0, 1.0))
echo f
```

- `macros.newLit` now works for ref object types.
- `macro pragmas` can now be used in type sections.
- 5 new pragmas were added to Nim in order to make the upcoming tooling more
  convenient to use. Nim compiler checks these pragmas for syntax but otherwise
  ignores them. The pragmas are `requires`, `ensures`, `assume`, `assert`, `invariant`.
- `system.writeFile` has been overloaded to also support `openarray[byte]`.
- `asyncdispatch.drain` now properly takes into account `selector.hasPendingOperations`
  and only returns once all pending async operations are guaranteed to have completed.
- `sequtils.zip` now returns a sequence of anonymous tuples i.e. those tuples
  now do not have fields named "a" and "b".
- `distinctBase` has been moved from `sugar` to `typetraits` and now it is
  implemented as compiler type trait instead of macro. `distinctBase` in sugar
  module is now deprecated.
- `CountTable.mget` has been removed from `tables.nim`. It didn't work, and it
  was an oversight to be included in v1.0.
- `tables.merge(CountTable, CountTable): CountTable` has been removed.
  It didn't work well together with the existing inplace version of the same proc
  (`tables.merge(var CountTable, CountTable)`).
  It was an oversight to be included in v1.0.
- `asyncdispatch.drain` now consistently uses the passed timeout value for all
  iterations of the event loop, and not just the first iteration.
  This is more consistent with the other asyncdispatch APIs, and allows
  `asyncdispatch.drain` to be more efficient.
- `base64.encode` and `base64.decode` were made faster by about 50%.
- `htmlgen` adds [MathML](https://wikipedia.org/wiki/MathML) support
  (ISO 40314).
- `macros.eqIdent` is now invariant to export markers and backtick quotes.
- `htmlgen.html` allows `lang` in the `<html>` tag and common valid attributes.
- `macros.basename` and `basename=` got support for `PragmaExpr`,
  so that an expression like `MyEnum {.pure.}` is handled correctly.
- `httpclient.maxredirects` changed from `int` to `Natural`, because negative values
  serve no purpose whatsoever.
- `httpclient.newHttpClient` and `httpclient.newAsyncHttpClient` added `headers`
  argument to set initial HTTP Headers, instead of a hardcoded empty `newHttpHeader()`.
- `parseutils.parseUntil` has now a different behaviour if the `until` parameter is
  empty. This was required for intuitive behaviour of the strscans module
  (see bug #13605).
- `strutils.formatFloat` with `precision = 0` has the same behavior in all
  backends, and it is compatible with Python's behavior,
  e.g. `formatFloat(3.14159, precision = 0)` is now `3`, not `3.`.
- `times.parse` now only uses input to compute its result, and not `now`:
  `parse("2020", "YYYY", utc())` is now `2020-01-01T00:00:00Z` instead of
  `2020-03-02T00:00:00Z` if run on 03-02; it also doesn't crash anymore when
  used on 29th, 30th, 31st of each month.
- `httpcore.==(string, HttpCode)` is now deprecated due to lack of practical
  usage. The `$` operator can be used to obtain the string form of `HttpCode`
  for comparison if desired.
- `std/oswalkdir` was buggy, it's now deprecated and reuses `std/os` procs.
- `net.newContext` now performs SSL Certificate checking on Linux and OSX.
  Define `nimDisableCertificateValidation` to disable it globally.
- `os.walkDir` and `os.walkDirRec` now have new flag, `checkDir` (default: false).
  If it is set to true, it will throw if input dir is invalid instead of a noop
  (which is the default behaviour, as it was before this change),
  `os.walkDirRec` only throws if top-level dir is invalid, but ignores errors for
  subdirs, otherwise it would be impossible to resume iteration.
- The `FD` variant of `selector.unregister` for `ioselector_epoll` and
  `ioselector_select` now properly handle the `Event.User` select event type.
- `joinPath` path normalization when `/` is the first argument works correctly:
  `assert "/" / "/a" == "/a"`. Fixed the edge case: `assert "" / "" == ""`.
- `xmltree` now adds indentation consistently to child nodes for any number
  of children nodes.
- `os.splitPath()` behavior synchronized with `os.splitFile()` to return "/"
  as the dir component of `/root_sub_dir` instead of the empty string.
- The deprecated `lc` macro has been removed from `sugar`. It is now replaced with the
  more powerful `collect` macro.
- `os.relativePath("foo", "foo")` is now `"."`, not `""`, as `""` means invalid
  path and shouldn't be conflated with `"."`; use `-d:nimOldRelativePathBehavior`
  to restore the old behavior.
- `os.joinPath(a, b)` now honors trailing slashes in `b` (or `a` if `b` = "").
- `base64.encode` no longer supports `lineLen` and `newLine`.
  Use `base64.encodeMime` instead.


## Language changes

- An `align` pragma can now be used for variables and object fields, similar
  to the `alignas` declaration modifier in C/C++.
- `=sink` type bound operator is now optional. Compiler can now use combination
  of `=destroy` and `copyMem` to move objects efficiently.
- Unsigned integer operators have been fixed to allow promotion of the first operand.
- Conversions to unsigned integers are unchecked at runtime, imitating earlier Nim
  versions. The documentation was improved to acknowledge this special case.
  See nim-lang/RFCs#175 for more details.
- New syntax for lvalue references: `var b {.byaddr.} = expr` enabled by
  `import std/decls`.
- `var a {.foo.}: MyType = expr` now lowers to `foo(a, MyType, expr)` for
  non-builtin pragmas, enabling things like lvalue references (see `decls.byaddr`).


## Compiler changes

- Generated JS code uses spaces, instead of mixed spaces and tabs.
- The Nim compiler now supports the ``--asm`` command option for easier
  inspection of the produced assembler code.
- The Nim compiler now supports a new pragma called ``.localPassc`` to
  pass specific compiler options to the C(++) backend for the C(++) file
  that was produced from the current Nim module.
- The compiler now inferes "sink parameters". To disable this for a specific routine,
  annotate it with `.nosinks`. To disable it for a section of code, use
  `{.push sinkInference: off.}`...`{.pop.}`.
- The compiler now supports a new switch `--panics:on` that turns runtime
  errors like `IndexError` or `OverflowError` into fatal errors that **cannot**
  be caught via Nim's `try` statement. `--panics:on` can improve the
  runtime efficiency and code size of your program significantly.
- The compiler now warns about inheriting directly from `system.Exception` as
  this is **very bad** style. You should inherit from `ValueError`, `IOError`,
  `OSError` or from a different specific exception type that inherits from
  `CatchableError` and cannot be confused with a `Defect`.
- The error reporting for Nim's effect system has been improved.
- Implicit conversions for `const` behave correctly now, meaning that code like
  `const SOMECONST = 0.int; procThatTakesInt32(SOMECONST)` will be illegal now.
  Simply write `const SOMECONST = 0` instead.
- The `{.dynlib.}` pragma is now required for exporting symbols when making
  shared objects on POSIX and macOS, which make it consistent with the behavior
  on Windows.
- The compiler is now more strict about type conversions concerning proc
  types: Type conversions cannot be used to hide `.raise` effects or side
  effects, instead a `cast` must be used. With the flag `--useVersion:1.0` the
  old behaviour is emulated.
- The Nim compiler now implements a faster way to detect overflows based
  on GCC's `__builtin_sadd_overflow` family of functions. (Clang also
  supports these). Some versions of GCC lack this feature and unfortunately
  we cannot detect this case reliably. So if you get compilation errors like
  "undefined reference to `__builtin_saddll_overflow`" compile your programs
  with `-d:nimEmulateOverflowChecks`.


## Tool changes

- Nimpretty doesn't accept negative indentation argument anymore, because it was
  breaking files.


## Bugfixes

- Fixed "`nimgrep --nocolor` is ignored on posix; should be instead: `--nimgrep --color=[auto]|true|false`"
  ([#7591](nim-lang/Nim#7591))
- Fixed "Runtime index on const array (of converted obj) causes C-compiler error"
  ([#10514](nim-lang/Nim#10514))
- Fixed "windows x86 with vcc compile error with "asmNoStackFrame""
  ([#12298](nim-lang/Nim#12298))
- Fixed "[TODO] regression: Error: Locks requires --threads:on option"
  ([#12330](nim-lang/Nim#12330))
- Fixed "Add --cc option to --help or --fullhelp output"
  ([#12010](nim-lang/Nim#12010))
- Fixed "questionable `csize` definition in `system.nim`"
  ([#12187](nim-lang/Nim#12187))
- Fixed "os.getAppFilename() returns incorrect results on OpenBSD"
  ([#12389](nim-lang/Nim#12389))
- Fixed "HashSet[uint64] slow insertion depending on values"
  ([#11764](nim-lang/Nim#11764))
- Fixed "Differences between compiling 'classic call syntax' vs 'method call syntax' ."
  ([#12453](nim-lang/Nim#12453))
- Fixed "c -d:nodejs --> SIGSEGV: Illegal storage access"
  ([#12502](nim-lang/Nim#12502))
- Fixed "Closure iterator crashes on --newruntime due to "dangling references""
  ([#12443](nim-lang/Nim#12443))
- Fixed "No `=destroy` for elements of closure environments other than for latest devel --gc:destructors"
  ([#12577](nim-lang/Nim#12577))
- Fixed "strutils:formatBiggestFloat() gives different results in JS mode"
  ([#8242](nim-lang/Nim#8242))
- Fixed "Regression (devel): the new `csize_t` definition isn't consistently used, nor tested thoroughly..."
  ([#12597](nim-lang/Nim#12597))
- Fixed "tables.take() is defined only for `Table` and missed for other table containers"
  ([#12519](nim-lang/Nim#12519))
- Fixed "`pthread_key_t` errors on OpenBSD"
  ([#12135](nim-lang/Nim#12135))
- Fixed "newruntime: simple seq pop at ct results in compile error"
  ([#12644](nim-lang/Nim#12644))
- Fixed "[Windows] finish.exe C:\Users\<USERNAME>\.nimble\bin is not in your PATH environment variable."
  ([#12319](nim-lang/Nim#12319))
- Fixed "Error with strformat + asyncdispatch + const"
  ([#12612](nim-lang/Nim#12612))
- Fixed "MultipartData needs $"
  ([#11863](nim-lang/Nim#11863))
- Fixed "Nim stdlib style issues with --styleCheck:error"
  ([#12687](nim-lang/Nim#12687))
- Fixed "new $nimbleDir path substitution yields unexpected search paths"
  ([#12767](nim-lang/Nim#12767))
- Fixed "Regression: inlined procs now get multiple rounds of destructor injection"
  ([#12766](nim-lang/Nim#12766))
- Fixed "newruntime: compiler generates defective code"
  ([#12669](nim-lang/Nim#12669))
- Fixed "broken windows modules path handling because of 'os.relativePath' breaking changes"
  ([#12734](nim-lang/Nim#12734))
- Fixed "for loop tuple syntax not rendered correctly"
  ([#12740](nim-lang/Nim#12740))
- Fixed "Crash when trying to use `type.name[0]`"
  ([#12804](nim-lang/Nim#12804))
- Fixed "Enums should be considered Trivial types in Atomics"
  ([#12812](nim-lang/Nim#12812))
- Fixed "Produce static/const initializations for variables when possible"
  ([#12216](nim-lang/Nim#12216))
- Fixed "Assigning descriminator field leads to internal assert with --gc:destructors"
  ([#12821](nim-lang/Nim#12821))
- Fixed "nimsuggest `use` command does not return all instances of symbol"
  ([#12832](nim-lang/Nim#12832))
- Fixed "@[] is a problem for --gc:destructors"
  ([#12820](nim-lang/Nim#12820))
- Fixed "Codegen ICE in allPathsAsgnResult"
  ([#12827](nim-lang/Nim#12827))
- Fixed "seq[Object with ref and destructor type] doesn't work in old runtime"
  ([#12882](nim-lang/Nim#12882))
- Fixed "Destructor not invoked because it is instantiated too late, old runtime"
  ([#12883](nim-lang/Nim#12883))
- Fixed "The collect macro does not handle if/case correctly"
  ([#12874](nim-lang/Nim#12874))
- Fixed "allow typed/untyped params in magic procs (even if not in stdlib)"
  ([#12911](nim-lang/Nim#12911))
- Fixed "ARC/newruntime memory corruption"
  ([#12899](nim-lang/Nim#12899))
- Fixed "tasyncclosestall.nim still flaky test: Address already in use"
  ([#12919](nim-lang/Nim#12919))
- Fixed "newruntime and computed goto: variables inside the loop are in generated code uninitialised"
  ([#12785](nim-lang/Nim#12785))
- Fixed "osx: dsymutil needs to be called for debug builds to keep debug info"
  ([#12735](nim-lang/Nim#12735))
- Fixed "codegen ICE with ref objects, gc:destructors"
  ([#12826](nim-lang/Nim#12826))
- Fixed "mutable iterator cannot yield named tuples"
  ([#12945](nim-lang/Nim#12945))
- Fixed "parsecfg stores "\r\n" line breaks just as "\n""
  ([#12970](nim-lang/Nim#12970))
- Fixed "db_postgres.getValue issues warning when no rows found"
  ([#12973](nim-lang/Nim#12973))
- Fixed "ARC: Unpacking tuple with seq causes segfault"
  ([#12989](nim-lang/Nim#12989))
- Fixed "ARC/newruntime: strutils.join on seq with only empty strings causes segfault"
  ([#12965](nim-lang/Nim#12965))
- Fixed "regression (1.0.4): `{.push exportc.}` wrongly affects generic instantiations, causing codegen errors"
  ([#12985](nim-lang/Nim#12985))
- Fixed "cdt, crash with --gc:arc, works fine with default gc"
  ([#12978](nim-lang/Nim#12978))
- Fixed "ARC: No indexError thrown on out-of-bound seq access, SIGSEGV instead"
  ([#12961](nim-lang/Nim#12961))
- Fixed "ARC/async: Returning in a try-block results in wrong codegen"
  ([#12956](nim-lang/Nim#12956))
- Fixed "asm keyword is generating wrong output C code when --cc:tcc"
  ([#12988](nim-lang/Nim#12988))
- Fixed "Destructor not invoked"
  ([#13026](nim-lang/Nim#13026))
- Fixed "ARC/newruntime: Adding inherited var ref object to seq with base type causes segfault"
  ([#12964](nim-lang/Nim#12964))
- Fixed "Style check error with JS compiler target"
  ([#13032](nim-lang/Nim#13032))
- Fixed "regression(1.0.4): undeclared identifier: 'readLines'; plus another regression and bug"
  ([#13013](nim-lang/Nim#13013))
- Fixed "regression(1.04) `invalid pragma: since` with nim js"
  ([#12996](nim-lang/Nim#12996))
- Fixed "Sink to MemMove optimization in injectdestructors"
  ([#13002](nim-lang/Nim#13002))
- Fixed "--gc:arc: `catch` doesn't work with exception subclassing"
  ([#13072](nim-lang/Nim#13072))
- Fixed "nim c --gc:arc --exceptions:{setjmp,goto} incorrectly handles raise; `nim cpp --gc:arc` is ok"
  ([#13070](nim-lang/Nim#13070))
- Fixed "typetraits feature request - get subtype of a generic type"
  ([#6454](nim-lang/Nim#6454))
- Fixed "CountTable inconsistencies between keys() and len() after setting value to 0"
  ([#12813](nim-lang/Nim#12813))
- Fixed "{.align.} pragma is not applied if there is a generic field"
  ([#13122](nim-lang/Nim#13122))
- Fixed "ARC, finalizer, allow rebinding the same function multiple times"
  ([#13112](nim-lang/Nim#13112))
- Fixed "`nim doc` treats `export localSymbol` incorrectly"
  ([#13100](nim-lang/Nim#13100))
- Fixed "--gc:arc SIGSEGV (double free?)"
  ([#13119](nim-lang/Nim#13119))
- Fixed "codegen bug with arc"
  ([#13105](nim-lang/Nim#13105))
- Fixed "symbols not defined in the grammar"
  ([#10665](nim-lang/Nim#10665))
- Fixed "[JS] Move is not defined"
  ([#9674](nim-lang/Nim#9674))
- Fixed "[TODO] pathutils.`/` can return invalid AbsoluteFile"
  ([#13121](nim-lang/Nim#13121))
- Fixed "regression(1.04) `nim doc main.nim` generates broken html (no css)"
  ([#12998](nim-lang/Nim#12998))
- Fixed "Wrong supportsCopyMem on string in type section"
  ([#13095](nim-lang/Nim#13095))
- Fixed "Arc, finalizer, out of memory"
  ([#13157](nim-lang/Nim#13157))
- Fixed "`--genscript` messes up nimcache and future nim invocations"
  ([#13144](nim-lang/Nim#13144))
- Fixed "--gc:arc with --exceptions:goto for "nim c" generate invalid c code"
  ([#13186](nim-lang/Nim#13186))
- Fixed "[regression] duplicate member `_i1` codegen bug"
  ([#13195](nim-lang/Nim#13195))
- Fixed "RTree investigations with valgrind for --gc:arc"
  ([#13110](nim-lang/Nim#13110))
- Fixed "relativePath("foo", ".") returns wrong path"
  ([#13211](nim-lang/Nim#13211))
- Fixed "asyncftpclient - problem with welcome.msg"
  ([#4684](nim-lang/Nim#4684))
- Fixed "Unclear error message, lowest priority"
  ([#13256](nim-lang/Nim#13256))
- Fixed "Channel messages are corrupted"
  ([#13219](nim-lang/Nim#13219))
- Fixed "Codegen bug with exportc and case objects"
  ([#13281](nim-lang/Nim#13281))
- Fixed "[bugfix] fix #11590: c compiler warnings silently ignored, giving undefined behavior"
  ([#11591](nim-lang/Nim#11591))
- Fixed "[CI] tnetdial flaky test"
  ([#13132](nim-lang/Nim#13132))
- Fixed "Cross-Compiling with -d:mingw fails to locate compiler under OSX"
  ([#10717](nim-lang/Nim#10717))
- Fixed "`nim doc --project` broken with imports below main project file or duplicate names"
  ([#13150](nim-lang/Nim#13150))
- Fixed "regression: isNamedTuple(MyGenericTuple[int]) is false, should be true"
  ([#13349](nim-lang/Nim#13349))
- Fixed "--gc:arc codegen bug copying objects bound to C structs with missing C struct fields"
  ([#13269](nim-lang/Nim#13269))
- Fixed "write requires conversion to string"
  ([#13182](nim-lang/Nim#13182))
- Fixed "Some remarks to stdlib documentation"
  ([#13352](nim-lang/Nim#13352))
- Fixed "a `check` in unittest generated by template doesn't show actual value"
  ([#6736](nim-lang/Nim#6736))
- Fixed "Implicit return with case expression fails with 'var' return."
  ([#3339](nim-lang/Nim#3339))
- Fixed "Segfault with closure on arc"
  ([#13314](nim-lang/Nim#13314))
- Fixed "[Macro] Crash on malformed case statement with multiple else"
  ([#13255](nim-lang/Nim#13255))
- Fixed "regression: `echo 'discard' | nim c -r -` generates a file '-' ; `-` should be treated specially"
  ([#13374](nim-lang/Nim#13374))
- Fixed "on OSX, debugging (w gdb or lldb) a nim program crashes at the 1st call to `execCmdEx`"
  ([#9634](nim-lang/Nim#9634))
- Fixed "Internal error in getTypeDescAux"
  ([#13378](nim-lang/Nim#13378))
- Fixed "gc:arc mode breaks tuple let"
  ([#13368](nim-lang/Nim#13368))
- Fixed "Nim compiler hangs for certain C/C++ compiler errors"
  ([#8648](nim-lang/Nim#8648))
- Fixed "htmlgen does not support `data-*` attributes"
  ([#13444](nim-lang/Nim#13444))
- Fixed "[gc:arc] setLen will cause string not to be null-terminated."
  ([#13457](nim-lang/Nim#13457))
- Fixed "joinPath("", "") is "/" ; should be """
  ([#13455](nim-lang/Nim#13455))
- Fixed "[CI] flaky test on windows: tests/osproc/texitcode.nim"
  ([#13449](nim-lang/Nim#13449))
- Fixed "Casting to float32 on NimVM is broken"
  ([#13479](nim-lang/Nim#13479))
- Fixed "`--hints:off` doesn't work (doesn't override ~/.config/nim.cfg)"
  ([#8312](nim-lang/Nim#8312))
- Fixed "joinPath("", "") is "/" ; should be """
  ([#13455](nim-lang/Nim#13455))
- Fixed "tables.values is broken"
  ([#13496](nim-lang/Nim#13496))
- Fixed "global user config can override project specific config"
  ([#9405](nim-lang/Nim#9405))
- Fixed "Non deterministic macros and id consistency problem"
  ([#12627](nim-lang/Nim#12627))
- Fixed "try expression doesn't work with return on expect branch"
  ([#13490](nim-lang/Nim#13490))
- Fixed "CI will break every 4 years on feb 28: times doesn't handle leap years properly"
  ([#13543](nim-lang/Nim#13543))
- Fixed "[minor] `nimgrep --word` doesn't work with operators (eg misses  `1 +% 2`)"
  ([#13528](nim-lang/Nim#13528))
- Fixed "`as` is usable as infix operator but its existence and precedence are not documented"
  ([#13409](nim-lang/Nim#13409))
- Fixed "JSON unmarshalling drops seq's items"
  ([#13531](nim-lang/Nim#13531))
- Fixed "os.joinPath returns wrong path when head ends '\' or '/' and tail starts '..'."
  ([#13579](nim-lang/Nim#13579))
- Fixed "Block-local types with the same name lead to bad codegen (sighashes regression)"
  ([#5170](nim-lang/Nim#5170))
- Fixed "tuple codegen error"
  ([#12704](nim-lang/Nim#12704))
- Fixed "newHttpHeaders does not accept repeated headers"
  ([#13573](nim-lang/Nim#13573))
- Fixed "regression: --incremental:on fails on simplest example"
  ([#13319](nim-lang/Nim#13319))
- Fixed "strscan can't get value of last element in format"
  ([#13605](nim-lang/Nim#13605))
- Fixed "hashes_examples crashes with "Bus Error" (unaligned access) on sparc64"
  ([#12508](nim-lang/Nim#12508))
- Fixed "gc:arc bug with re-used `seq[T]`"
  ([#13596](nim-lang/Nim#13596))
- Fixed "`raise CatchableError` is broken with --gc:arc  when throwing inside a proc"
  ([#13599](nim-lang/Nim#13599))
- Fixed "cpp --gc:arc --exceptions:goto fails to raise with discard"
  ([#13436](nim-lang/Nim#13436))
- Fixed "terminal doesn't compile with -d:useWinAnsi"
  ([#13607](nim-lang/Nim#13607))
- Fixed "Parsing "sink ptr T" - region needs to be an object type"
  ([#12757](nim-lang/Nim#12757))
- Fixed "gc:arc + threads:on + closures compilation error"
  ([#13519](nim-lang/Nim#13519))
- Fixed "[ARC] segmentation fault"
  ([#13240](nim-lang/Nim#13240))
- Fixed "times.toDateTime buggy on 29th, 30th and 31th of each month"
  ([#13558](nim-lang/Nim#13558))
- Fixed "Deque misbehaves on VM"
  ([#13310](nim-lang/Nim#13310))
- Fixed "Nimscript listFiles should throw exception when path is not found"
  ([#12676](nim-lang/Nim#12676))
- Fixed "koch boot fails if even an empty config.nims is present in ~/.config/nims/ [devel regression]"
  ([#13633](nim-lang/Nim#13633))
- Fixed "nim doc generates lots of false positive LockLevel warnings"
  ([#13218](nim-lang/Nim#13218))
- Fixed "Arrays are passed by copy to iterators, causing crashes, unnecessary allocations and slowdowns"
  ([#12747](nim-lang/Nim#12747))
- Fixed "Range types always uses signed integer as a base type"
  ([#13646](nim-lang/Nim#13646))
- Fixed "Generate c code cannot compile with recent devel version"
  ([#13645](nim-lang/Nim#13645))
- Fixed "[regression] VM: Error: cannot convert -1 to uint64"
  ([#13661](nim-lang/Nim#13661))
- Fixed "Spurious raiseException(Exception) detected"
  ([#13654](nim-lang/Nim#13654))
- Fixed "gc:arc memory leak"
  ([#13659](nim-lang/Nim#13659))
- Fixed "Error: cannot convert -1 to uint (inside tuples)"
  ([#13671](nim-lang/Nim#13671))
- Fixed "strformat issue with --gc:arc"
  ([#13622](nim-lang/Nim#13622))
- Fixed "astToStr doesn't work inside generics"
  ([#13524](nim-lang/Nim#13524))
- Fixed "oswalkdir.walkDirRec wont return folders"
  ([#11458](nim-lang/Nim#11458))
- Fixed "`echo 'echo 1' | nim c -r -`  silently gives wrong results (nimBetterRun not updated for stdin)"
  ([#13412](nim-lang/Nim#13412))
- Fixed "gc:arc destroys the global variable accidentally."
  ([#13691](nim-lang/Nim#13691))
- Fixed "[minor] sigmatch errors should be sorted, for reproducible errors"
  ([#13538](nim-lang/Nim#13538))
- Fixed "Exception when converting csize to clong"
  ([#13698](nim-lang/Nim#13698))
- Fixed "ARC: variables are no copied on the thread spawn causing crashes"
  ([#13708](nim-lang/Nim#13708))
- Fixed "Illegal distinct seq causes compiler crash"
  ([#13720](nim-lang/Nim#13720))
- Fixed "cyclic seq definition crashes the compiler"
  ([#13715](nim-lang/Nim#13715))
- Fixed "Iterator with openArray parameter make the argument evaluated many times"
  ([#13417](nim-lang/Nim#13417))
- Fixed "net/asyncnet: Unable to access peer's certificate chain"
  ([#13299](nim-lang/Nim#13299))
- Fixed "Accidentally "SIGSEGV: Illegal storage access" error after arc optimizations (#13325)"
  ([#13709](nim-lang/Nim#13709))
- Fixed "Base64 Regression"
  ([#13722](nim-lang/Nim#13722))
- Fixed "A regression (?) with --gc:arc and repr"
  ([#13731](nim-lang/Nim#13731))
- Fixed "Internal compiler error when using the new variable pragmas"
  ([#13737](nim-lang/Nim#13737))
- Fixed "bool conversion produces vcc 2019 warning at cpp compilation stage"
  ([#13744](nim-lang/Nim#13744))
- Fixed "Compiler "does not detect" a type recursion error in the wrong code, remaining frozen"
  ([#13763](nim-lang/Nim#13763))
- Fixed "[minor] regression: `Foo[0.0] is Foo[-0.0]` is now false"
  ([#13730](nim-lang/Nim#13730))
- Fixed "`nim doc` - only whitespace on first line causes segfault"
  ([#13631](nim-lang/Nim#13631))
- Fixed "hashset regression"
  ([#13794](nim-lang/Nim#13794))
- Fixed "`os.getApplFreebsd` could return incorrect paths in the case of a long path"
  ([#13806](nim-lang/Nim#13806))
- Fixed "Destructors are not inherited"
  ([#13810](nim-lang/Nim#13810))
- Fixed "io.readLines AssertionError on devel"
  ([#13829](nim-lang/Nim#13829))
- Fixed "exceptions:goto accidentally reset the variable during exception handling"
  ([#13782](nim-lang/Nim#13782))
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

6 participants