Version 0.18.0 released
2018年03月01日 The Nim Team
The Nim team is happy to announce that the latest release of Nim, version 0.18.0, is now available. Nim is a systems programming language that focuses on performance, portability and expressiveness.
Installing 0.18.0
If you have installed a previous version of Nim using choosenim,
getting Nim 0.18.0 is as easy as:
$ choosenim update stable
If you don’t have it already, you can get choosenim by following
these instructions or you can install
Nim manually by following the instructions on our
install page.
Note that Windows users might still have an easier time with the
unzip+finish.exe method of installing Nim which is described on the
install page.
What’s new in 0.18.0
It’s been a while since our last release, but we’ve been busy. At over 1000 commits since the last release, v0.18.0 is one of (if not the) biggest releases of Nim ever.
We have mentioned multiple times that this will be a major release. Primarily because our goal has been to clean up the standard library significantly in preparation for v1.0. We’ve made massive progress in this release to achieve that goal. The downside is that this release has a much larger proportion of breaking changes than normal releases, but it is all in the spirit of a cleaner Nim v1.0.
Breaking changes to be mindful of
You will find that the [] indexing operator which takes a slice now raises
an IndexError exception when it is out of bounds. Previously it would have
simply returned the part of the string that was captured. You may find that your
program hasn’t taken this into account and will now crash with an IndexError
exception. To get back the previous behaviour, use the substr procedure.
Here is an example of the issue and how to get around it:
var myString = "hello world"
doAssertRaises IndexError:
discard myString[6 .. 45]
doAssert myString.substr(6, 45) == "world"
The $ operator now handles collections differently. For example:
doAssert $(@["23", "12"]) == "@[\"23\", \"12\"]"
# Whereas in 0.17.2:
doAssert $(@["23", "12"]) == "@[23, 12]"
An array[x, char] can no longer be converted to a cstring type. This
has the side-effect of allowing $ to be defined on an array:
var x: array[5, char]
doAssert $x == r"['\x00', '\x00', '\x00', '\x00', '\x00']"
Be sure to check out the changelog below for a comprehensive list of breaking changes. If you run into any strange regressions feel free to pop into IRC, Gitter, Discord or any of the other chat rooms/forums we have available. Links to them all are available in our community page.
New features
Of course, this being a major new release, there are also plenty of new features.
Formatted string literals
The greatest new addition is the strformat module. It implements string
formatting that is very similar to Python 3’s f-strings.
Formatted string literals are prefixed with fmt or the & operator:
import strformat
let name = "Fred"
let age = 24
let weight = 94.52
doAssert fmt"My name is {name}. I'm {age} years of age and weigh {weight:.3f}." ==
"My name is Fred. I'm 24 years of age and weigh 94.520."
For more information, take a look at the
strformat documentation.
Testable documentation examples
The documentation generator can now test your examples for you! This is done
using the new runnableExamples macro.
proc add*(x, y: int): int =
## Adds two numbers together.
##
## Examples:
##
runnableExamples:
doAssert add(5, 5) == 10
doAssert add(-5, 2) == -3
x + y
If you save this as addLib.nim and then generate its documentation via
nim doc addLib.nim, you should see something like this:
Hint: /tmp/addLib_examples [Exec]
Hint: operation successful (13129 lines compiled; 0.492 sec total; 19.742MiB peakmem; Debug Build) [SuccessX]
The code under runnableExamples will be embedded in the procedure’s
documentation.
If your examples have an error in them then you will see it, together with a stack trace if your error occurs at runtime. This is great for ensuring that your documentation is up-to-date.
mapLiterals
This is a new macro that allows you to easily create array and sequence literals.
import sequtils
let x = mapLiterals([12, 34, 15, 1], uint32)
doAssert x is array[4, uint32]
New memory manager algorithm
The new TLSF algorithm has been implemented to reduce memory fragmentation.
This has a side effect of making alloc and dealloc O(1) operations.
Async improvements
There have been a number of various improvements to the asynchronous IO modules in the stdlib. There is no longer an “upcoming” and standard asyncdispatch split, the former was merged and is now the default asyncdispatch implementation.
A new getIoHandler procedure is now available, it returns a handle
to the underlying IO completion port or epoll/kqueue fd used by asyncdispatch.
Main advantage of this is that libraries now have even greater control over the
event loop.
As an example, it allows much more efficient HTTP server implementations
(one such implementation is called httpbeast).
A new implementation of async await has also been added for the JavaScript
backend in the asyncjs module.
Nimble v0.8.10
In the release are also updates to the Nim package manager. The latest version of Nimble is included and contains multiple new features and bug fixes.
The major new feature is the support for multiple Nimble packages in a single Git/Hg repository.
Check out the changelog for a list of changes since Nimble’s last release.
Contributors to v0.18.0
Our contributors are amazing, and there is far too many to list here. Big thanks to all of you, we couldn’t have pulled off this release without you!
Changelog
Changes affecting backwards compatibility
Breaking changes in the standard library
-
The
[]proc for strings now raises anIndexErrorexception when the specified slice is out of bounds. See issue #6223 for more details. You can usesubstr(str, start, finish)to get the old behaviour back, see this commit for an example. -
strutils.splitandstrutils.rsplitwith an empty string and a separator now returns that empty string. See issue #4377. -
Arrays of char cannot be converted to
cstringanymore, pointers to arrays of char can! This means$for arrays can finally exist insystem.nimand do the right thing. This means$myArrayOfCharchanged its behaviour! Compile with-d:nimNoArrayToStringto see where to fix your code. -
reExtendedis no longer default for thereconstructor in theremodule. -
The behavior of
$has been changed for all standard library collections. The collection-to-string implementations now perform proper quoting and escaping of strings and chars. -
newAsyncSockettaking anAsyncFDnow runssetBlocking(false)on the fd. -
modand bitwiseanddo not producerangesubtypes anymore. This turned out to be more harmful than helpful and the language is simpler without this special typing rule. -
formatFloat/formatBiggestFloatnow support formatting floats with zero precision digits. The previousprecision = 0behavior (default formatting) is now available viaprecision = -1. - Moved from stdlib into Nimble packages:
basic2ddeprecated: useglm,arraymancer,neo, or another package insteadbasic3ddeprecated: useglm,arraymancer,neo, or another package insteadgentabslibuvpolynumericpdcursesromanslibsvmjoyent_http_parser
-
Proc toCountTable now produces a
CountTablewith values corresponding to the number of occurrences of the key in the input. It used to produce a table with all values set to1.Counting occurrences in a sequence used to be:
let mySeq = @[1, 2, 1, 3, 1, 4] var myCounter = initCountTable[int]() for item in mySeq: myCounter.inc itemNow, you can simply do:
let mySeq = @[1, 2, 1, 3, 1, 4] myCounter = mySeq.toCountTable() -
If you use
--dynlibOverride:sslwith OpenSSL 1.0.x, you now have to defineopenssl10symbol (-d:openssl10). By default OpenSSL 1.1.x is assumed. -
newNativeSocketis now namedcreateNativeSocket. -
newAsyncNativeSocketis now namedcreateAsyncNativeSocketand it no longer raises an OS error but returns anosInvalidSocketwhen creation fails. -
The
securehashmodule is now deprecated. Instead importstd / sha1. - The
readPasswordFromStdinproc has been moved from therdstdinto theterminalmodule, thus it does not depend on linenoise anymore.
Breaking changes in the compiler
-
\nis now only the single line feed character like in most other programming languages. The new platform specific newline escape sequence is written as\p. This change only affects the Windows platform. -
The overloading rules changed slightly so that constrained generics are preferred over unconstrained generics. (Bug #6526)
-
We changed how array accesses “from backwards” like
a[^1]ora[0..^1]are implemented. These are now implemented purely insystem.nimwithout compiler support. There is a new “heterogeneous” slice typesystem.HSlicethat takes 2 generic parameters which can beBackwardsIndexindices.BackwardsIndexis produced bysystem.^. This means if you overload[]or[]=you need to ensure they also work withsystem.BackwardsIndex(if applicable for the accessors). -
The parsing rules of
ifexpressions were changed so that multiple statements are allowed in the branches. We found few code examples that now fail because of this change, but here is one:t[ti] = if exp_negative: '-' else: '+'; inc(ti)This now needs to be written as:
t[ti] = (if exp_negative: '-' else: '+'); inc(ti) -
The experimental overloading of the dot
.operators now take anuntypedparameter as the field name, it used to be astatic[string]. You can usewhen defined(nimNewDot)to make your code work with both old and new Nim versions. See special-operators for more information. -
yield(orawaitwhich is mapped toyield) never worked reliably in an array, seq or object constructor and is now prevented at compile-time.
Library additions
-
Added
sequtils.mapLiteralsfor easier construction of array and tuple literals. -
Added
system.runnableExamplesto make examples in Nim’s documentation easier to write and test. The examples are tested as the last step ofnim doc. -
Implemented
getIoHandlerproc in theasyncdispatchmodule that allows you to retrieve the underlying IO Completion Port orSelector[AsyncData]object in the specified dispatcher. -
For string formatting / interpolation a new module called strformat has been added to the stdlib.
-
The
ReadyKeytype in the selectors module now contains anerrorCodefield to help distinguish betweenEvent.Errorevents. -
Implemented an
acceptproc that works on aSocketHandleinnativesockets. -
Added
algorithm.rotateLeft. -
Added
typetraits.$as an alias fortypetraits.name. -
Added
system.getStackTraceEntriesthat allows you to access the stack trace in a structured manner without string parsing. -
Added
parseutils.parseSaturatedNatural. -
Added
macros.unpackVarargs. -
Added support for asynchronous programming for the JavaScript backend using the
asyncjsmodule. -
Added true color support for some terminals. Example:
import colors, terminal const Nim = "Efficient and expressive programming." var fg = colYellow bg = colBlue int = 1.0 enableTrueColors() for i in 1..15: styledEcho bgColor, bg, fgColor, fg, Nim, resetStyle int -= 0.01 fg = intensity(fg, int) setForegroundColor colRed setBackgroundColor colGreen styledEcho "Red on Green.", resetStyle
Library changes
-
echonow works with strings that contain\0(the binary zero is not shown) andnilstrings are equal to empty strings. -
JSON: Deprecated
getBVal,getFNum, andgetNumin favour ofgetBool,getFloat,getBiggestInt. A newgetIntprocedure was also added. -
rationals.toRationalnow uses an algorithm based on continued fractions. This means its results are more precise and it can’t run into an infinite loop anymore. -
os.getEnvnow takes an optionaldefaultparameter that tellsgetEnvwhat to return if the environment variable does not exist. -
The
randomprocs inrandom.nimhave all been deprecated. Instead use the newrandprocs. The module now exports the state of the random number generator as typeRandso multiple threads can easily use their own random number generators that do not require locking. For more information about this rename see issue #6934 -
writeStackTraceis now proclaimed to have no IO effect (even though it does) so that it is more useful for debugging purposes. -
db_mysqlmodule:DbConnis now adistincttype that doesn’t expose the details of the underlyingPMySQLtype. -
parseopt2is now deprecated, useparseoptinstead.
Language additions
-
It is now possible to forward declare object types so that mutually recursive types can be created across module boundaries. See package level objects for more information.
-
Added support for casting between integers of same bitsize in VM (compile time and nimscript). This allows to, among other things, reinterpret signed integers as unsigned.
-
Custom pragmas are now supported using pragma
pragma, please see language manual for details. -
Standard library modules can now also be imported via the
stdpseudo-directory. This is useful in order to distinguish between standard library and nimble package imports:import std / [strutils, os, osproc] import someNimblePackage / [strutils, os]
Language changes
-
The unary
<is now deprecated, for.. <use..<for other usages use thepredproc. -
Bodies of
forloops now get their own scope:# now compiles: for i in 0..4: let i = i + 1 echo i -
To make Nim even more robust the system iterators
..andcountupnow only accept a single generic typeT. This means the following code doesn’t die with an “out of range” error anymore:var b = 5.Natural var a = -5 for i in a..b: echo i -
atomicandgenericare no longer keywords in Nim.genericused to be an alias forconcept,atomicwas not used for anything. -
The memory manager now uses a variant of the TLSF algorithm that has much better memory fragmentation behaviour. According to http://www.gii.upv.es/tlsf/ the maximum fragmentation measured is lower than 25%. As a nice bonus
allocanddeallocbecame O(1) operations. -
The compiler is now more consistent in its treatment of ambiguous symbols: Types that shadow procs and vice versa are marked as ambiguous (bug #6693).
-
codegenDecl pragma now works for the JavaScript backend. It returns an empty string for function return type placeholders.
-
Extra semantic checks for procs with noreturn pragma: return type is not allowed, statements after call to noreturn procs are no longer allowed.
- Noreturn proc calls and raising exceptions branches are now skipped during common type
deduction in
ifandcaseexpressions. The following code snippets now compile:import strutils let str = "Y" let a = case str: of "Y": true of "N": false else: raise newException(ValueError, "Invalid boolean") let b = case str: of nil, "": raise newException(ValueError, "Invalid boolean") elif str.startsWith("Y"): true elif str.startsWith("N"): false else: false let c = if str == "Y": true elif str == "N": false else: echo "invalid bool" quit("this is the end") -
Pragmas now support call syntax, for example:
{.exportc"myname".}and{.exportc("myname").} -
The
deprecatedpragma now supports a user-definable warning message for procs.proc bar {.deprecated: "use foo instead".} = return bar()
Tool changes
-
The
nim doccommand is now an alias fornim doc2, the second version of the documentation generator. The old version 1 can still be accessed via the newnim doc0command. -
Nim’s
rst2htmlcommand now supports the testing of code snippets via an RST extension that we called:test:::.. code-block:: nim :test: # shows how the 'if' statement works if true: echo "yes"
Bugfixes
- Fixed “ReraiseError when using try/except within finally block” (#5871)
- Fixed “Range type inference leads to counter-intuitive behvaiour” (#5854)
- Fixed “JSON % operator can fail in extern procs with dynamic types” (#6385)
- Fixed ““intVal is not accessible” in VM” (#6083)
- Fixed “Add excl for OrderedSet” (#2467)
- Fixed “newSeqOfCap actually doesn’t reserve memory” (#6403)
- Fixed “[Regression] Nim segfaults” (#6435)
- Fixed “Seq assignment is slower than expected” (#6433)
- Fixed “json module issues with empty dicts and lists” (#6438)
- Fixed “mingw installed via finish.exe fails to link if Nim located in path with whitespace” (#6452)
- Fixed “unittest.check does not perform short-circuit evaluation” (#5784)
- Fixed “Error while concatenating an array of chars.” (#5861)
- Fixed “range initialization: [ProveInit] hint: Cannot prove that” (#6474)
- Fixed “scanf can call procs with side-effects multiple times” (#6487)
- Fixed “gcsafe detection problem” (#5620)
- Fixed “C++ codegen:
mitemsgenerates invalid code.” (#4910) - Fixed “strange runtime behavior on macOS” (#6496)
- Fixed “stdtmpl: invalid indentation after a line ending in question mark” (#5070)
- Fixed “Windows: NAN troubles on c backend” (#6511)
- Fixed “lib/nim/system/cellsets.nim(33, 31) Error: type mismatch while attempting to compile for 16bit CPUs” (#3558)
- Fixed “Can’t compile dynlib with
-d:useNimRtland--threads:on” (#5143) - Fixed “var s = @[0,1,2,…] can generate thousand of single assignments in C code” (#5007)
- Fixed “
echodiscards everything after a null character” (#1137) - Fixed “Turn off reExtended by default” (#5627)
- Fixed “Bad Links in docs/backends.html” (#5914)
- Fixed “Index out of bounds error in db_postgres when executing non parameter-substituted queries containing “?”” (#6571)
- Fixed “Please add pipe2 support to posix stdlib” (#6553)
- Fixed “Return semantics vary depending on return style” (#6422)
- Fixed “parsecsv.open reports SIGSEGV when calling ‘open’ on missing file” (#6148)
- Fixed “VCC: Nim generates non-compilable code for system.nim” (#6606)
- Fixed “Generic subtype matches worse than a generic” (#6526)
- Fixed “formatFloat inconsistent scientific notation” (#6589)
- Fixed “Generated c code calls function twice” (#6292)
- Fixed “Range type inference leads to counter-intuitive behvaiour” (#5854)
- Fixed “New backward indexing is too limited” (#6631)
- Fixed “Table usage in a macro (SIGSEGV: Illegal storage access.)” (#1860)
- Fixed “Incorrect deprecation error” (#6634)
- Fixed “Wrong indices in arrays not starting with 0” (#6675)
- Fixed “if expressions” (#6609)
- Fixed “BackwardsIndex: converter +
[]+ unrelated type[^1]: lib/system.nim(3536, 3) Error” (#6692) - Fixed “js backend 0.17.3: array bounds check for non zero based arrays is buggy” (#6532)
- Fixed “HttpClient’s new API doesn’t work through a proxy for https URLs” (#6685)
- Fixed “isServing isn’t declared and isn’t compiling” (#6707)
-
Fixed “[Regression] value out of range” (#6710)
-
Fixed “Error when using
multisyncmacro” (#6708) - Fixed “formatFloat inconsistent scientific notation” (#6589)
- Fixed “Using : (constructor arguments) for passing values to functions with default arguments causes a compiler crash.” (#6765)
- Fixed “In-place object initialization leads to vcc incompatible code” (#6757)
- Fixed “Improve parseCookies doc” (#5721)
- Fixed “Parser regression with nested do notation inside conditional” (#6166)
- Fixed “Request for better error message” (#6776)
- Fixed “Testament tester does not execute test with
exitcodeonly” (#6775) - Fixed “JS integer division off by one” (#6753)
- Fixed “Regression: cannot prove not nil” (#5781)
- Fixed “SIGSEGV: Illegal storage access. (Attempt to read from nil?) in generic proc” (#6073)
- Fixed “Request for better error message” (#6776)
- Fixed “Nim #head: sorting via reference hangs compiler” (#6724)
- Fixed “Cannot cast pointer to char in cpp” (#5979)
- Fixed “asynchttpserver replies with several errors on single request” (#6386)
- Fixed “object variants superclass trigger bad codegen” (#5521)
- Fixed “JS integer division off by one” (#6753)
- Fixed “js backend compiler crash with tables indexed by certain types” (#6568)
- Fixed “Jsgen bug with is” (#6445)
- Fixed “Subrange definition with ..<” (#6788)
- Fixed “fields not initialized: array with enum index type as object field.” (#6682)
- Fixed “Can not delete data in table when table’s data type is kind of “not nil”” (#6555)
- Fixed “tables.nim: Cannot prove that ‘n’ is initialized” (#6121)
- Fixed “issues with ‘not nil’ applied to a closure proc” (#6489)
- Fixed “
not nilnot working in some cases” (#4686) -
Fixed “Cannot prove ‘@[v]’ is not nil” (#3993)
- Fixed “Feature: support TCP_NODELAY in net.sockets” (#6795)
- Fixed “Code that makes the compiler throw an error message and then hangs” (#6820)
- Fixed “Code that makes the compiler throw an error message and then hangs” (#6820)
- Fixed “Inconsistent behavior with sequence and string slicing” (#6223)
- Fixed “Wrong behavior of “split” (proc and iterator)” (#4377)
- Fixed “[Documentation] Invalid module name: [foo, bar]” (#6831)
- Fixed “The destructor is not called for temporary objects” (#4214)
- Fixed “Destructors does not work with implicit items iterator in for loop” (#985)
- Fixed “Error in template when using the type of the parameter inside it” (#6756)
- Fixed “should json.to() respect parent attributes?” (#5856)
- Fixed “json ‘to’ macro can not marshalize into tuples” (#6095)
- Fixed “json.to fails with seq[T]” (#6604)
- Fixed “json.to() is not worth using compared to marshal.to[T]” (#5848)
- Fixed “Memory not being released in time, running out of memory” (#6031)
- Fixed “[Regression] Bad C codegen for generic code” (#6889)
- Fixed “rollingFileLogger deletes file on every start.” (#6264)
- Fixed “Remove/deprecate securehash module.” (#6033)
- Fixed “[bug or not] object construction for seq[T] failed without space after colon” (#5999)
- Fixed “issues with the random module” (#4726)
- Fixed “Reassigning local var to seq of objects results in nil element in Object’s seq field” (#668)
- Fixed “Compilation error with “newseq[string]”” (#6726)
- Fixed “await inside array/dict literal produces invalid code - Part 2” (#6626)
- Fixed “terminal.eraseline() gives OverflowError on Windows” (#6931)
- Fixed “[Regression]
sequtils.anyconflicts withsystem.any” (#6932) - Fixed “C++ codegen:
mitemsgenerates invalid code.” (#4910) - Fixed “seq.mitems produces invalid cpp codegen” (#6892)
- Fixed “Concepts regression” (#6108)
- Fixed “Generic iterable concept with array crashes compiler” (#6277)
- Fixed “C code generation “‘a’ is a pointer; did you mean to use ‘->’?”” (#6462)
- Fixed “
--NimblePathfails if a-in path which is not followed by a number” (#6949) - Fixed ““not registered in the selector” in asyncfile.close() for something that clearly was registered” (#6906)
-
Fixed “strange frexp behavior” (#6353)
- Fixed “noreturn branches of case statements shouldn’t contribute to type” (#6885)
- Fixed “Type inference for ‘if’ statements changed” (#6980)
- Fixed “newest asyncdispatch recursion” (#6100)
-
Fixed “Ambiguous identifier between set type and proc” (#6965)
-
Fixed “Inconsistent behavior with sequence and string slicing” (#6223)
- Fixed “Unsupported OpenSSL library imported dynamically” (#5000)
- Fixed “
nim checksegfaults” (#6972) - Fixed “GC deadlock” (#6988)
- Fixed “Create a seq without memory initialization” (#6401)
- Fixed “Fix bug for getch on Windows while using the arrow keys” (#6966)
- Fixed “runnableExamples doesn’t work in templates” (#7018)
-
Fixed “runnableExamples doesn’t work with untyped statement blocks” (#7019)
- Fixed “Critical bug in parseBiggestFloat” (#7060)
- Fixed “[RFC] strformat.% should be gone” (#7078)
- Fixed “compiler crash on simple macro” (#7093)
- Fixed “Make newlines sane again” (#7089)
- Fixed “JS - Unicode enum string representation issue” (#6741)
- Fixed “Strange behaviour of 0.17.3 (working ok in 0.17.2)” (#6989)
- Fixed “Strange behaviour of 0.17.3 (working ok in 0.17.2)” (#6989)
- Fixed “Compiler crash: try expression with infix as” (#7116)
- Fixed “nimsuggest crash” (#7140)
- Fixed “[RFC] Reintroduce readChar” (#7072)
- Fixed “Copyright line needs updating” (#7129)
- Fixed “-0.0 doesn’t result in negative zero in VM” (#7079)
- Fixed “Windows large filesize” (#7121)
- Fixed “Securehash is not parsimonious with MD5 and other hash modules” (#6961)
- Fixed “os.findExe() shouldn’t look in current directory on posix, unless exe has a /” (#6939)
- Fixed “
compiles(...)withfatalpragma causes compiler to exit early” (#7080) - Fixed “NPE when compile macro that returns concrete value” (#5450)
- Fixed “Using a variable of type
int | floatcauses internal compiler error” (#6946) - Fixed “Unsigned integers could not be used as array indexes.” (#7153)
- Fixed “countdown with uint causes underflow” (#4220)
- Fixed “Inconsistent method call syntax” (#7200)
- Fixed “Impossible to create an empty const array” (#6853)
- Fixed “Strange UINT handling” (#3985)
- Fixed “Bad codegen when passing arg that is part of return value destination” (#6960)
- Fixed “No info about gcsafety in error message when global var is accessed in async proc” (#6186)
- Fixed “Resolving package vs. local import ambiguities” (#2819)
- Fixed “Internal error with type() operator” (#3711)
- Fixed “newAsyncSocket should raise an OS error plus other inconsistencies” (#4995)
- Fixed “mapLiterals fails with negative values” (#7215)
- Fixed “fmWrite doesn’t truncate file with openAsync, unlike open()” (#5531)
- Fixed “Move libsvm to an external nimble module” (#5786)
- Fixed “Prevent acceptAddr gotcha with newSocket” (#7227)
-
Fixed “strtabs.getOrDefault is inconsistent with tables.getOrDefault” (#4265)
- Fixed “Code falling through into exception handler when no exception thrown.” (#7232)
- Fixed “the new generic inference rules are broken” (#7247)
- Fixed “Odd
..<regression” (#6992) - Fixed “Different proc type inferred from default parameter” (#4659)
- Fixed “Different proc type inferred from default parameter” (#4659)
- Fixed “Testament sometimes ignores test failures” (#7236)
- Fixed “New Allocator Fails On >=4GB Requests” (#7120)
- Fixed “User pragmas hide effect specifications from sempass2” (#7216)
- Fixed “C++: SIGABRT instead of IndexError for out-of-bounds” (#6512)
- Fixed “An uncaught exception in cpp mode doesn’t show the exception name/msg” (#6431)