[prev in list] [next in list] [prev in thread] [next in thread]
List: llvm-bugs
Subject: [llvm-bugs] [Bug 55790] `noderef` attribute does not work in `[[]]` spelling
From: LLVM Bugs via llvm-bugs <llvm-bugs () lists ! llvm ! org>
Date: 2022-05-31 7:17:07
Message-ID: 20220531071707.9065d8186d08ed14 () email ! llvm ! org
[Download RAW message or body]
[Attachment #2 (text/html)]
<table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Issue</th>
<td>
<a href=https://github.com/llvm/llvm-project/issues/55790>55790</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>
`noderef` attribute does not work in `[[]]` spelling
</td>
</tr>
<tr>
<th>Labels</th>
<td>
clang:frontend
</td>
</tr>
<tr>
<th>Assignees</th>
<td>
</td>
</tr>
<tr>
<th>Reporter</th>
<td>
martinboehme
</td>
</tr>
</table>
<pre>
While working on https://reviews.llvm.org/D126061, I noticed that the `noderef` \
attribute does not work in the `[[]]` spelling.
Here is an example ([godbolt link](https://godbolt.org/z/f19Yszoxn)), with comments \
describing the actual behavior as well as what I think the desired behavior should \
be:
```
void f() {
// Actual: warning: 'noderef' can only be used on an array or pointer \
type // Desired: Either keep current diagnostic or forbid `noderef` on declarations
[[clang::noderef]] int i;
// Actual: No diagnostic
// Desired: Either diagnose the faulty dereference or forbid `noderef` on \
declarations [[clang::noderef]] int *p1;
*p1;
// Actual: No diagnostic
// Desired: Either diagnose the faulty dereference or forbid `noderef` on \
declarations int *p2 [[clang::noderef]];
*p2;
// Actual: error: 'noderef' attribute cannot be applied to types
// Desired: Allow attribute in this position, but warn that it can only be used \
on array or pointer types int [[clang::noderef]] *p3;
*p3;
// Actual: warning: 'noderef' can only be used on an array or pointer \
type // Desired: Allow attribute in this position, diagnose the faulty dereference
int * [[clang::noderef]] p4;
*p4;
// Actual: warning: 'noderef' can only be used on an array or pointer \
type [-Wignored-attributes] // Desired: Either
// - diagnose the faulty dereference
// - forbid `noderef` on typedefs, or
// - forbid `noderef` on declarations
typedef int *IntPtr;
[[clang::noderef]] IntPtr p5;
*p5;
// Actual: error: 'noderef' attribute cannot be applied to types
// Desired: Either diagnose the faulty dereference or forbid `noderef` on \
typedefs typedef int *IntPtr;
IntPtr [[clang::noderef]] p6;
*p6;
}
```
It's not exactly clear to me why the attribute doesn't work in the `[[]]` spelling. \
From my understanding of the code, my expectation would be that the declaration \
`[[clang::noderef]] int *p1;` should cause a warning on the subsequent dereference \
(even though it's not really correct to place a `[[]]` type attribute on a \
declaration). In debugging this, I have verified that a `NoDerefAttr` does get placed \
on the `int` type, just like it does with the `__attribute__` spelling. From there on \
out, the different spelling should not really have an influence on the behavior, but \
apparently they do; I have not been able to determine the reason for this.
Certainly, one thing that makes `noderef` special is that there is special logic in \
SemaType.cpp to associate `noderef` with the pointer operand to the _right_ of the \
attribute, rather than the pointer operand to the _left_ of the attribute (see \
`TypeProcessingState::parsedNoDeref` and `ExpectNoDerefChunk` in \
`GetFullTypeForDeclarator()`). Correspondingly, the `NoDerefAttr` is placed on the \
pointee type, rather than the pointer type, as would be more usual for pointer \
attributes. This doesn't explain why the `[[]]` spelling doesn't work though.
This special behavior will however make it more difficult to get the `[[]]` spelling \
to work correctly. The C++ standard has clear rules on the entity that an attribute \
appertains to in a particular position, and attributes that appertain to a pointer \
must be placed to the right of the `*`. A `noderef` pointer should therefore be \
declared as `int * [[clang::noderef]] p4;`. This implies that the `NoDerefAttr` would \
be placed on the pointer type, rather than the pointee type as it is done today.
Probably the best way to fix this would be to retain the special processing for the \
GNU spelling of the attribute, but modify it so that the the `NoDerefAttr` is placed \
on the pointer type; I believe the existing `ExpectNoDerefChunk` logic, which is \
currently only used for diagnostics, could be used to do this. The code that \
diagnoses dereferences of `noderef` pointers would then be changed correspondingly to \
expect the attribute on the pointer, not the pointee.
As https://reviews.llvm.org/D126061 will make changes to the way `[[]]` attributes \
are handled, I would request that any fix should wait until that patch has landed.
</pre>
<img width="1px" height="1px" alt="" \
src="http://email.email.llvm.org/o/eJzNWE1v2zgQ_TXyhYhhK3EcH3zIR9PNpSjQLhZ7CmiJstlQopa \
k4ri_ft8MJdmO3cTFortbqEksUcOZN2_eDL2w-Wb-x0obJdbWPelqKWwlViHUPjm_TtJ7XE49a7X2Q2Oey6F1S \
9y6G6eXo8txkt6KB1HZoDOVi7CSAT-USC5Hlc2VUwX-EjIEpxdNUCK3ytNq3kroqlucTG74uqMLb_haGQNXhsn \
oLhldx5-_wZ7QXshKqBdZ1vA4Sa_w3tLmC2uCwBtPZCC92ne_fd56_h3_i_HsT__dvlRJOuPrVqx1WInMlqWqg \
he58hl8JjTIRZmFRhqxUCv5rK0T0os1POTfFPMDVmFzXotXtQMY_WK_so2hz-TQTjwUd7z447PVuSgoonQmkul \
NvCvwL0YhrtkJGBFr6Sq4Rn8m6bRDOp2KDNjYymywmWg8nEAqcUs6JzcCrtRWV0E5ETa1OrB_Fx0nqx8ABpY9K \
VWLrHEOmIhcy2VlPTJNlgrrFvB3P9HYLVeZkU4GbSu_swPnF0_YaVzdS5xxECEInZzf7KJzNPBPdseNEwJoFyt \
OTCEbEzaCN8b_KlO_JJAkva7HfTDRvd07_-fwOv_TdwI9CC89JTzlnHWHpN3KA-hL4gDyyro2mhTFMlf9W2BcG \
2PXO1ZYVqATtfWa4qPqxgOumihROhyvlGNl8hqdtwlAWJwfoHN-Cjq_vKhPwekdRh0w5T086osDMC7-NTDIubM \
_NAICAmd94J58e7e4DlacnQ5O_8aPqo-8y1XhCXJ7bKufK9vWXJeUhyp8Dm4f-LfTFN8Q9eQgW5P_sLD_ucr1O \
P8EVC0W7xH78gCq7Z1kene0xcefDyH5kCZXo2Q2i8MQppksgNKZUdIRMiVmsdUmTh57s1PVv3n6CCXunS1FuRF \
NhQB8kFXOQ17Bb2aIiliI5-qlVllgbsF4HFm2I90O87bbndYIyZs4A2USNStkV96cItj2zcKrvxqeMnYSi1lIP \
StaYpvlCqr9CjenpCHYLAaULBBwtZEZ2T-CBwvCFk2Sjd2YMHQNkXrcWjTLZZz7tI_jLeY4JZ6V04Xuhlze4pO \
9I2-vYZR24OF2qUL0Iu-CwyOg0blAFr81nobVJ0WNiN_i4bNd_PjYe_n4eCSTVBPsv20CWePk6IJBC_3iDvEdn \
DgMaKWuCtPEyokedpNq1yhRo5KMGaYgis0iix0OsY6RFbnAAA7McwW9LXUV6xN7eZhFQTKAewP8rXJBaog3yx6 \
_EIEGoKV8Ag77BYxYMo2xGz2qo2E8AXQPjF1iGkUNfFGl_Ap0h1ldk0_Se4sV4fVJpMe5axO2Vg4FwXKE249OL \
1fhsSuOPhHkMHhCagRPqjdNGFUcsUBk9or9IUc_O5sp7xH-F1SciiUE1NHRWlbxwaliTfvAhdnev101OOXgoeY \
6_KjCfWMM2by37q4lNOWSzhFUB0TsWyoRX1su_Yh_y7ZXHKZxYI-9MUrVc_dHKHTP6UTUaUeJvosmTSenYqc1b \
xvxUHylAeS1skGIDHjSa-Ab6nZcFaNg7HGPN-p40x_N1hqHuJVdQ2YcU5Aqkt2mitIZeg3llYr6PUewjPdu1ch \
sKDglbpP0Bpdg3ZUuRxH5VuhdY0D5FmeUmw6bVlyqHd6gGGPZeNpCk2yBJ4F8k25vdiO2bLFtTXVvc1X0KShJg \
ZChNtctdZn8HXOZOtTAhuL6VRV1VlqJ4bIsCLRF1yZgU_pW-U4fE2kvzpMuaUjYVv0RovYkO0ZX9w5dVdsNPKW \
bCUhiZHO52eMMinQBldu0IunpCLEhsAr9EkfnbZu0UL6IM_Wzlmd1X-WtICrx8dPvW84ckxkS4NKCfhtyztstC \
D9VsS0ErNsLBTSfowH1onG4xN4_1BUWVf5GZKWzFdlvvwEAEDx589hN8WzPqtwosw4MXkCdwcYewIVAg0aMpZv \
n_G6z9wTGUZp1KMP7iqxnyOUSG2T7mkYbxgnmlfDuw0KOUgvbIcNezq_9yd99RfVg1Yg--a6QiCZHtGKnOFEiU \
IIqN5h0ecaIMTqagnzoZGDDTGvLbC3BhwYyYeLjWgakh-QEVZWrvA1jkM_P89n5TA4gKEbNT_4q7g1xGzTOzF9 \
9p4ZO2iyGmS3xgbBpf52B89-QBHzU3iMY_DGZTGejwWo-mqYX-biQciGvAODk_HIyyhYqH08up2M5LgZGgqp-T \
j6kaScWhbPIUgWcUrg10PN0lKajyfl4NB1PR9PhDJbyq_HVZT66gq2L5GKEaUCbPmMDN2fPMNh5PDQogG06BxgVcEJUah4jH8gG3cPNS9LYamHVqlQDjmTOYfwN-rzZKQ">
[Attachment #3 (text/plain)]
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic