Contracts nonattr rebase onto r16 4626 gd9641b64e988 #158
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
At present, for pre-conditions and for post-conditions with a void return, we are not emitting a return statement. This patch adds the relevant return statements. gcc/cp/ChangeLog: * contracts.cc (finish_function_contracts): Add return statements to pre-condition and void post-cndition checking functions. Signed-off-by: Iain Sandoe <[email protected]>
Since we now have only void functions for pre and post checks, we can simplify the code that syntesizes the function decls.
P2680 shows a situation where optimization of contracts expressions that exploit assumption that UB cannot occur - by completely eliding checks in those cases; which actually prevents the contract checking to work as the user intended. As a proof-of-principle, this patch allows use to disable optimization of the contract checks. It is not proposed as a production solution - really we need to figure out the proper course of action.
… in the correct place in the grammar. Attributes on contracts not implemented yet.
…_chainon handles error_mark_nodes properly) - disabled inheriting contracts from base classes on virtual functions - removed allowing contracts to appear on non first declarations - removed check in for fcontracts-nonattr as it doesn't seem needed, but introduces an issue in diagnosing misapplied contracts - renamed the 'new' test suite to P2900
'build_over_call' can return a TARGET_EXPR which the contracts wrapper machinery is not expecting. 1. move the wrapper insertion into build_cxx_call, so that we can do it before the TARGET_EXPRs are generated (which is easier than doing surgery on the generated ones). 2. Do not copy the original fn RESULT_DECL, but instead create a new one that has the correct type for the wrapper return (I am not sure, but suspect that the original fn return type can have been altered by copy elision). 3. The hack is this: - in build_cxx_call, we check the fndecl to see how to handle it. - this is done using 'get_callee_fndecl()' which does not understand how to look through virtual calls.. and therefore returns NULL_TREE which means we then fail to add the wrapper. - probably, the Right Fix is to make a cp_get_callee_fndecl() that does understand how to look through virtual calls.. - but, instead since there is provision to pass in an 'orig_fndecl' and we actually know that in build_over_call, I added some logic to do that. I added contract test case for the object return and a trivial coroutines one. Not sure whether any of this is working properly with optimsation enabled (possibly because of the known D_A_O issue) Signed-off-by: Iain Sandoe <[email protected]>
not generating duplicate definitions of pre and post check fixing early mangling for checked function fixing the context of postcondition identifier of a wrapper function
For a non-template member function of a class template, the PARM_DECL is't TREE_READONLY, but the type of the parameter is const. Add handling for that case.
We have altered the ordering of some of the code-gen and now it is possible to encounter ignored and quick_enforce cases in get_contract_semantic(). Signed-off-by: Iain Sandoe <[email protected]>
We wee looking up __cxa_current_exception_object in the global namespace but the function (even tho it's extern "C") lives in __cxxabi. Signed-off-by: Iain Sandoe <[email protected]>
Start work on de-attribution of P2900 contracts; make it explicit when we are using attributes. Signed-off-by: Iain Sandoe <[email protected]>
More renaming to be explicit about attributes, make the arguments to the contract matcher contracts rather than attributes containing contracts. Signed-off-by: Iain Sandoe <[email protected]>
Some general tidies, move the use of attributes one step up the call stack in the parser. Signed-off-by: Iain Sandoe <[email protected]>
They are not attributes, and treating them as such is hindering progress. In the revised design, they are kept as on-the-side data for function decls that have contract specifiers (in a similar way to other out-of- band specifiers). Since we are currently still supporting attribute-based contracts for the cxx2a case, and there is some share code, in the first instance we have a number of cases wherewe have to select the source of the data depending on flag_contracts_nonattr. Eventually, these cases will be removed (1) because we want to do the non-attribute work a different way and (2) because we will remove the support for the attribute case which was not adopted in the WD. Signed-off-by: Iain Sandoe <[email protected]>
We have, at present, a mixture of pieces of declatations in cp-tree.h
and contracts.h (which is also included in cp-tree.h). Remove all the
contracts information from cp-tree.h and re-organise it in contracts.h.
Then include contracts.h where it is needed only.
In a similar manner we have two implementations in contracts.cc where
one of them (cxx2a contracts) is intended to be deprecated and finally
removed. However there is currently shared support code and common
public APIs. This reorganises the code with the following head comment.
"
To facilitate maintenance, please keep the general ordering of content:
shared local support functions
cxx2a implementation:
local support functions
cxx2a-only public API
cxx26 implementation:
local support functions
cxx26-only public API
shared public API.
"
In the process, we identified some APIs that are actually local to
contracts.cc and these have been made static functions and removed
from the public API.
Signed-off-by: Iain Sandoe <[email protected]>
This replaces any uses of the to-be-deprecated attribute format options and instead uses the current C++26 versions. Small tidy-ups in unused options and duplicated std specs. Signed-off-by: Iain Sandoe <[email protected]>
When we have a specialisation, it might be given contract specifiers (depending on whether they are present on the template, as a side-effect of the instantiation process). However, these added specifiers are then deleted - since the presence of (and spelling of) specifiers on specialisations is independent of the original template. We were doing this by adding empty contracts to the specialisation. This has an unfortunate side-effect of adding decls to the specifier lookup map, even if the decl had no contracts already. This can play badly with GC and hash collisions. Solve this by adding an interface that only deletes specifiers on functions already registered. This is subtely different from the case when we are about to ggc_delete a decl (in that case we actually also want to remove the decl from the map too). So renamed that interface to say that it will remove the decl. Signed-off-by: Iain Sandoe <[email protected]>
We still have the option to outline them with -fcontract-checks-outlined. Switch the tests that were checking -fno-contract-checks-outlined to do the opposite. Signed-off-by: Iain Sandoe <[email protected]>
This does all the work needed to have the builtin type but does not (yet) publish it - since we are still figuring out the details of the ABI - however, In this version, we no longer try to declare or build a std::source_location since this creates quite some difficulty with needing the declaration whether <source_location> is included or not. Instead, we save a type-erased pointer to the implementation data. Since clang and GCC actually use the same layout here, there is a move to add this to the Itanium spec at that point we will be able to use some __cxa_xxxxx_t * as the pointer type. Signed-off-by: Iain Sandoe <[email protected]>
This allows us to put values for semantic and assertion type that will not fit into the LANG_ bits This is intended to allow the library to make use of compiler-side logic to implement changes in semantic handling, rather than duplicating this.
Since we want to accommodate semantics with a larger range of values than can be fitted into the TREE_LANG_xxx bits, we have made a slot in the contract expression tree to hold the semantic. This switches the P2900-based implementation to use that. Signed-off-by: Iain Sandoe <[email protected]>
Now we can accommodate values in the recommended practice range proposed by the WD, switch the noexcept_xxx semantics to use them. Signed-off-by: Iain Sandoe <[email protected]>
We want to allow contract checks in constructors and dsestructors to have controlled optimisations; one way to achieve this is to outline the checks and apply the constraints there. This was already possible for other functions. Co-authored-by: Nina Ranns <[email protected]> Signed-off-by: Iain Sandoe <[email protected]>
…dentifiers We had a situation in which post condition function names were being mangled with inclusion of the added return value parameter, this made it possible that we could get a duplicate function name if there was a void overload with an extra parm. Co-authored-by: Iain Sandoe <[email protected]> Signed-off-by: Iain Sandoe <[email protected]>
This alters the logic slightly in the violation_handler wrapper so that we only generate one call to terminate. Signed-off-by: Iain Sandoe <[email protected]>
This implements a wrapper around calls to std::terminate that are made for enforce and quick_enforce semantics. This works around GCC Bugzilla 121936 while a suitable fix is found for the upstream implementation. Co-authored-by: Iain Sandoe <[email protected]>
Signed-off-by: Iain Sandoe <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
...