[cfe-dev] ModuleCache issues on Linux (Ubuntu)

[Ilya Kuteev via cfe-dev]

I figured out that the problem is in reading PCM files using FileManager.
clang/lib/Basic/FileManager.cpp:271
```
FileEntry &UFE = UniqueRealFiles[Status.getUniqueID()];
```
If I remove this cache everything is OK.
Seems like inode is shared between deleted/edited and newly created PCM file.
If it happens FileManager returns a FileEntryRef to an invalid file.

How can I fix this issue? Maybe we need to remove UniqueRealFiles cache
for PCM files, or make it work with this case?


> 1 марта 2021 г., в 20:13, David Blaikie <dblaikie at gmail.com> написал(а):
> 
> Can't say any of this sounds especially familiar to me - cc'd a few folks who might have some context.
> 
> On Fri, Feb 19, 2021 at 6:28 AM Ilya Kuteev via cfe-dev <cfe-dev at lists.llvm.org <mailto:cfe-dev at lists.llvm.org>> wrote:
> Hello everyone!
> I've ran into a strange problem using clang-modules on Linux. I have a Linux machine running Ubuntu 18.04.4 LTS.
> When I put my ModuleCache on RAM everything is OK
> When I put ModuleCache on SSD/HDD, very frequently I'm getting different errors with pcm files, like below:
> ```
> fatal error: module file '/home/ilyakuteev/module_cache/2/1721172304169885905/3OA4HVRBFA8AJ/Base-5W2XL4QH4QAH.pcm' is out of date and needs to be rebuilt: signature mismatch
> ```
> 
> ```
> fatal error: malformed or corrupted AST file: 'Unexpected end of file reading 4154469704 of 4738416 bytes'
> note: after modifying system headers, please delete the module cache at '/home/ilyakuteev/module_cache/1721172304169885905/37WO4K6LHZNB5'
> fatal error: error in backend: Invalid abbrev number
> ```
> 
> ```
> fatal error: malformed or corrupted AST file: 'declaration ID out-of-range for AST file'
> note: after modifying system headers, please delete the module cache at '/home/ilyakuteev/module_cache/3/1721172304169885905/1907ERGA34IRD'
> ```
> 
> I’ve done some research and it looks like the failure is inside single clang process, clang writes pcm down on disk, then reads it and gets a malformed file back, or some other problem with reading on-disk pcm.
> 
> I've found a patch https://reviews.llvm.org/D22636 <https://reviews.llvm.org/D22636> and it looks like it fixes my problems, but I don’t want to use it, cause it just rebuilts "owned" module if its AST is corrupted and looks like retry of operation, not a real fix. Also this fix introduces some unstable behavior.
> 
> My questions:
> 1) Is this a known problem?
> 2) How can I safely hot-fix it before it will be fixed in clang?
> 3) Do I need to provide more logs for this issue to be fixed?
> 
> Thanks
> 
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at lists.llvm.org <mailto:cfe-dev at lists.llvm.org>
> https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev <https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20210301/34f85580/attachment.html>