add malloc_stats_clear routine to zero the heap master and all active thread heaps
(details)
Incorporated changes in response to Trevor's comments.
(details)
Added references when incorporating Trevor's comments
(details)
Added cfa equivalent of cofor/cobegin/coend
(details)
Added another check to the invariants for SizeofExpr/AlignofExpr.
(details)
committed makefile change for cofor include
(details)
Update max/min so n-ary version doesn't match against the 2 argument case. Also added a test.
(details)
Some mistakes in isPolyType found while working on the box pass. There was a new-ast verison using the old type map and, perhaps relatedly, the new type map version recursed into the no type map form. Also some clean-up in related functions.
(details)
add command setgcolumn to set lstlisting comment-column alignment, change size of lstlisting ^ character
(details)
general updates to the Cforall user manual
(details)
add keywords alignas, alignof, static_assert, and thread_local
(details)
Translated the box pass to the new AST. This includes direct as possible translations of the existing passes are two fix in passes which correct AST problems the direct translation causes. Outside the box pass there have already been many changes, ad there is another in Instantiate Generics, which disconnects designators instead of leaving them connected to the original polymorphic type, which breaks the invarants once the fields are removed in the Eraser sub-pass. There was also a change that was made and un-made in one commit. If translate from the new-AST to the old-AST part way you must, where possible, sort the TypeEnvKey values by string comparison. However, it now passes over that so it would be just extra complexity and run time, so I removed it. I stand at the exit from a great woods, just shy of a year from when I entered it. It has been a difficult and tiring journey. The path has been long and at times comically winding; but most often it was invisible, hidden under an impenetrable canopy and I spend days looking for it. All for a short jog forward before getting lost again. In front of me is another woods. It looks smaller, but I can't see the other side. Anyways, time to keep walking.
(details)
add corun/cofor statement, update old cofor
(details)
Translated the Link-Once pass to the new ast.
(details)
increase timeout in test.py from 180 to 240 seconds to deal with long locking tests on the ARM
(details)
First clean-up pass on box pass. Some of it is stuff that could have been done in the initial commit but there was just so much of it. The rest is mostly simple restructuring that makes the translation less direct.
(details)
Updated some documentation about the Instantiate Generic pass.
(details)
Implement "scanset" semantics for manipulator `skip`, in place of exact-match.
(details)
Support match-failure reporting for string input.
(details)
Fixes to Box pass's DeclAdapter. The two largest are: Only local information in scopeTypeVars was used so it was replaced with localTypeVars, that greatly simplified the pass. The added parameters are also divided into groups more consistently.
(details)
Fix read-to-variable-length-string cases when internal buffer fills.
(details)
Fix compiler bug where duplicate type declarations caused crash.
(details)
formatting, add PRNG copy to checkpoint PRNG state, remove autogen PRNG copy constructor and assignment to prevent accidental PRNG copy
(details)
change host from arm64 to aarch64 for ARM nightly build
(details)
Took the new EraseWith pass out of the box pass. It might be able to go even earlier or folded into an existing pass. C code generation now will not generate WithStmt nodes.
(details)
formatting, SKULLDUGGERY to remove spurious free-nonheap-object warning
(details)
Boxing no longer passes layout information about polymorphic types in the signature of the function. This simplifies the sub-passes in the box pass that added the arguments to the calls, the parameters to the declarations, and used the information internally. The effects in functions are less well quantified, because there are cases were the information has to be generated instead of passed in, but it is also never passed in (or generated) when not needed at all.
(details)
Translation of the fix main pass to the new ast data structures.
(details)
Updated some stats/counters to trigger off the new AST. Others will just have to be updated/re-added later because the support code to run them is missing and I don't know what stats we might need in the future.
(details)
Code dumps (under the -P flag) no longer convert to the old ast to print.
(details)
I don't actually know if this fixes anything, but the old version was definitely not doing anything and this is should have the same behaviour as the old one, and run just a bit faster.
(details)
Parser now uses constants from the new ast types.
(details)
Simple rework to BasicTypes-gen.cc to avoid refering to the old ast. No larger rework at this time.
(details)
Added a missing include (not sure how that slipped through) and did some other replacements so the new ast is the source of constants.
(details)
Hang on, eval doesn't have any side effects.
(details)
Took out some old code from the new section of the Mangler.
(details)
I forgot to move genName over to the new CodeGenerator module. It isn't really a code gen thing, more a modified lookup.
(details)
Changed some new uses of UniqueId to the ast version. Changed where some debugging variables are stored.
(details)
rename files gcc-builtins.cf builtins.cf extras.cf bootloader.cf and sync-builtins.cf with suffix .cfa
(details)
Added test for the checks on the constructor/destructor attributes.
(details)
removed print from static and dynamic tests that could be reordered due to a race
(details)
Refactored the OperatorTable module. Most is just an update in style and some documentation. I did add a function that should be a useful utility for the demangler.
(details)
Updated the demangler to also demangle operator names. You can also pass a filename to demangle now. There is also a test, but I don't really know what good tests are for the demangler.
(details)
removed fences for ARM that are likely not needed, insertion of fences into waituntil codegen TBD
(details)
forgot to include in last push. Added a get_proc_count that doesn't need a cluster argument
(details)
The demangler now uses the compiler's genType. The only difference I can find is that the demangler genType wrote functions ': RET_TYPE' instead of the C style. If this or a different change turns out to be imported it will have to be replicated post-translation.
(details)
Translated the demangling code from the old ast to the new ast.
(details)
Made UniqueName use other tools we have, removed the unused default and did some general clean-up.
(details)
Took the special main linkage code out of FunctionDecl and put it into a new pass. There is also a lot of related clean-up done.
(details)
input manipulators for getline, excl, and incl reset argument to null string for failed read, temporarily make reading a C-string have the same semantics
(details)
Remove the warning directive from SpecCost, you can't actually use the standard maybe_accept here, it does not allow the user to insert code between visiting elements in a container.
(details)
Fixed non-local ehm memory leak that occurs when a coroutine is deleted while suspended in a catchresume handler
(details)
added test case for non-local ehm memory leak
(details)
Some clean-up in src/Makefile.am (no change in behaviour).
(details)
It turns out enumerate does work if you use this very particular form of the for loop added in C++17. Ironically in this simple vector case it is about as complex. Swapped the order of fields in the enuemate value to make the binding read key-then-value.
(details)
Major round of clean-up in the GenPoly directory.
(details)
added architecture specific expect files for non-local exception test since it uses prng
(details)
added arm64 expect file for non-local exception test
(details)
added x86 expect file for non-local exception test cardgame
(details)
Round of clean-up in Tuples directory. (Skipping TupleExpansion, which will be fused together later.)
(details)
Reformating and clean-up in SymTab directory.
(details)
fixed bug where uninit mem was sometimes printed in test
(details)
slight restructuring of automake rules for generating the demangler
(details)
disable building the demangler (executable and library)
(details)
fixed bug where cofor runner could derefence a null pointer
(details)
Did some investigation of WithStmt. It may not be possible to convert it to a Stmt without changing how SymbolTable handles the with clauses.
(details)
Added a test showing how the conflict handling in the with clause currently works.
(details)
Removed Decl::fromId as it was unused. There are a few places that use uniqueId directly.
(details)
Round of significant clean-up and reindentation of InitTweak directory.
(details)
Round of clean-up in the Concurrency directory headers.
(details)
Box pass Eraser now removes the polymorphic arguments now that the functions are no longer actually polymorphic.
(details)
Slight improvement to forall clause erasing.
(details)
Removed unused case in CurrentObject::findNext. Did a bit of clean-up in the area.
(details)
Completed the second demangler/automake task: The demangler (when built) in played in the driver directory beside cfa-cc.
(details)
Removed the old-ast-compatable FunctionDecl constructor. However, enough cases pass nothing polymorphic along some of the uses of the constructor now go to a new monomorphic function constructor.
(details)
Updated .gitignore for the new location of demangler.
(details)
Combined the new and old TupleExpansion files.
(details)
Removed 'New' suffixes, they are no longer needed for disambiguation.
(details)
Renamed ResolvMode to ResolveMode. This is less consistent with the namespace, but is more consistent with almost everything else.
(details)
Rename the ScrubTyVars moduel to ScrubTypeVars. This makes it more consistent with the data types and function names now used.
(details)
Rename the linkReferenceTypes pass and the file that contains it. BaseInstType used to be called ReferenceToType, so the reason for the pass being called that is no longer true.
(details)
Removed some code that supported ast conversion, it supported ast translation, and that is done.
(details)
Implicit and layout parameters are now marked as const. This is primarily a safety feature to catch bad writes.
(details)
Just some random clean-up in Box pass while I was musing.
(details)
Some general clean-up, removing bits and pieces from the old version.
(details)
Moved include from Decl header. Removed some old ast code from the Parser.
(details)
Clean-up of the chain mutator. Seems like it is underused.
(details)
(Not really readable right now. Will clean up today and probably tomorrow) 1. Remove implicit conversion; 2. Companion trait;
(details)
Try terminate now does not call the catch function, now they have the same caller. This involved updating some platform dependent code which should be correct in all cases.
(details)
Accept Peter's proofreading and adjustment of examples to current syntax
(details)
Another attempt at fixing execptions. It is very close to the last attempt but the offsets have been updated and checked.
(details)
Clean-up in the exception translate pass. This changes some patterns from pre-translation and sets some things up for later reworks.
(details)
You can how use local control flow out of 'catch' clauses. Added a test to show that it works.
(details)
Forgot to remove an unused function from the exception code.
(details)
Reorganization of Linkage::Spec. is_mangled represented two properties with is_gcc_builtin separating them in one case. The second property is_overloadable which replaces is_gcc_builtin.
(details)
Small bug fix in code generation. Seems to only come up while generating intermediate code.
(details)
Remove extra compound statement and clean-up in InitTweak.
(details)
![[Jenkins]](/jenkins/static/96bd4631/images/svgs/logo.svg){kind=logo}
