[cfe-dev] Adding indexing support to Clangd

[Vladimir Voskresensky via cfe-dev]

On 15.08.2017 10:33, Alex L wrote:
>
>
> On 15 August 2017 at 07:57, Vladimir Voskresensky via cfe-dev 
> <cfe-dev at lists.llvm.org <mailto:cfe-dev at lists.llvm.org>> wrote:
>
>
>
>     On 14.08.2017 11:06, Ilya Biryukov wrote:
>>
>>         Could you provide more info, please.
>>         Are USRs different for same named file-local entries (i.e.
>>         static functions)?
>>
>>
>>     Looking at USRGenerator::VisitVarDecl, it seems they will be the
>>     same across different translation units.
>>     For local entities, "location" is part of the USR.  ("Location"
>>     is a filename + an offset from the start of that file).
>     That was my point, U in USRs are not so Unified, so can not be
>     used as universal keys for queries in DB.
>     Adding location to USR makes them less usable in index if fast
>     incremental re-index is going to be implemented vs. full reparse
>     on each change.
>
>     Vladimir.
>
>
> U is meant to be unique. USRs file-local declarations like static 
> functions will include the filename only. USRs for function-local 
> declarations will also include the offset in the file, but they're not 
> actually indexed. There are a couple of global declarations that 
> include file + offset in the USR as well - ObjC class extensions, some 
> embedded tag decls (not that common), template parameter names (not 
> actually indexed) & macros (not indexed).
>
> What exactly prevents unique strings from being used as keys for queries?
Keys have to be Unique.
And previously (in 3.6?) we observed that different entries had same 
USRs. Good to known it is fixed now.
>
> Are filenames are a problem for the incremental re-indexing? If they 
> aren't, then there are probably way to work around the file offsets in 
> USRs for the incremental re-indexing, especially given how uncommon 
> they actually are.
I just wanted to share our experience, that offsets added to unique 
strings are not good. They add extra complexity to invalidation phase 
during incremental re-indexing and not stable, i.e. during simple 
re-format of file or adding comments.
May be integral index per-file would be enough vs offset.

Vladimir.

>
> Alex
>
>>
>>     -- 
>>     Regards,
>>     Ilya Biryukov
>
>
>     _______________________________________________
>     cfe-dev mailing list
>     cfe-dev at lists.llvm.org <mailto:cfe-dev at lists.llvm.org>
>     http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>     <http://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/20170815/6c276653/attachment.html>