r342827 - Fix modules build with shared library.

Richard Smith via cfe-commits cfe-commits at lists.llvm.org
Mon Oct 22 16:42:53 PDT 2018 

On Mon, 22 Oct 2018 at 14:57, David Blaikie via cfe-commits <
cfe-commits at lists.llvm.org> wrote:

> Richard - any further thoughts here (re: layering/dependencies, etc)?
> Would love to get the layering oddity fixed rather than having it get more
> embedded over time.
>

Here's the intended current directory layout and layering as I understand
it:

 * include/clang/Analysis and lib/Analysis are the parts of the static
analysis engine that are depended on by both Sema and StaticAnalyzer, and
include common functionality / infrastructure
 * include/clang/Analysis/Analyses and lib/Analysis/Analyses are the
specific Sema analyses (that don't depend on the static analyzer);
StaticAnalyzer should not need to depend on this
 * the StaticAnalyzer library should contain all the pieces that are
specific to the static analyzer
 * Sema should not depend on Tooling, but it's fine for StaticAnalyzer to
depend on Tooling

Compared to where we are today, there are two differences:

1) The implementations of the headers in include/clang/Analysis are in
lib/Analysis, not in lib/Analysis/Analyses
2) ExprMutationAnalyzer is in lib/Analysis despite not being part of the
common infrastructure depended on by Sema and StaticAnalyzer

For point 1, we should change the lib/Analysis directory layout to match
the headers.
For point 2, we should find a home for this ExprMutationAnalyzer code that
makes sense. Given that the intent is to use it from the static analyzer,
that seems like a reasonable home for it, but libTooling would also make
some sense (perhaps a new subdirectory Tooling/Analysis), since it's only
performing AST matching, not a flow-sensitive / path-sensitive static
analysis.

On Tue, Oct 2, 2018 at 2:44 PM Shuai Wang via cfe-commits <
> cfe-commits at lists.llvm.org> wrote:
>
>> On Mon, Oct 1, 2018 at 4:58 PM Richard Smith <richard at metafoo.co.uk>
>> wrote:
>>
>>> On Mon, 1 Oct 2018 at 16:10, George Karpenkov via cfe-commits <
>>> cfe-commits at lists.llvm.org> wrote:
>>>
>>>> Hi Richard,
>>>>
>>>> On Oct 1, 2018, at 2:50 PM, Richard Smith via cfe-commits <
>>>> cfe-commits at lists.llvm.org> wrote:
>>>>
>>>> This looks like the wrong fix to me, but I don't really know enough
>>>> about what's being done with ExprMutationAnalyzer to have an opinion on
>>>> what the right fix is.
>>>>
>>>> Shuai, what is the goal here? Why is this code being moved to Analysis/?
>>>>
>>>>
>>>> I’ve asked for this possibility, as I wanted to use it from the Clang
>>>> static analyzer.
>>>>
>>>> Do you intend to call it from the compiler frontend at some point? I
>>>> can see a code review for the move itself, but no discussion of a plan or
>>>> overall direction being followed here. Did I miss it?
>>>>
>>>> We have historically decided to not use the tooling interfaces
>>>> (ASTMatcher, ParentMap, etc) from the frontend,
>>>>
>>>>
>>>> Clang analyzer uses ASTMatcher all over the place.
>>>>
>>>> because they're generally a poor fit for our goals (we aim for a
>>>> largely-single-pass compilation with good locality, and the AST matchers
>>>> make different design choices)
>>>>
>>>>
>>>> That’s totally good for the analyzer though, right?
>>>>
>>>
>>> Yes, that all seems fine for the static analyzer.
>>>
>>>
>>>> In any case, in future it might make sense to move the analyzer out of
>>>> Clang proper.
>>>> But for know the only way to use clang-tidy utilities from the analyzer
>>>> is to move them into Clang.
>>>>
>>>
>>> Right. I think we should try to maintain the idea that all the Clang
>>> Static Analyzer-specific things are in lib/StaticAnalyzer, and lib/Analysis
>>> only contains things depended on by the frontend. (That is: the things a
>>> clang binary would use if we didn't link in the static analyzer)
>>>
>>> Given the above, I think the best approach would be to move this code
>>> out of lib/Analysis and into somewhere in lib/StaticAnalyzer. There
>>> shouldn't be any problem with clang-tidy using it from there, since it
>>> already depends on the static analyzer.
>>>
>> I'm happy to do the move.
>> Could you (or someone) help point out where exactly I should move it to
>> though?
>>
>>> . If you want to change that, we'll need to discuss that decision.
>>>>
>>>> If the idea is to move this code into clang proper so that it can be
>>>> used by various different tooling clients, we'd need to discuss the right
>>>> place for it. Perhaps lib/Tooling/Analysis would make sense?
>>>>
>>>>
>>>> On Mon, 1 Oct 2018 at 13:13, David Blaikie via cfe-commits <
>>>> cfe-commits at lists.llvm.org> wrote:
>>>>
>>>>> I can't really reproduce this - when I try to build clang/llvm with
>>>>> LLVM_ENABLE_MODULES in CMake I'm still seeing an error I reported March on
>>>>> a cfe-dev thread - something to do with unique_ptr instantiations for
>>>>> MappedBlockStream in the PDB parsing code.
>>>>>
>>>>> So, I'm wondering what error you hit, Eric/where/how, etc...
>>>>>
>>>>> On Sun, Sep 30, 2018 at 10:23 AM Eric Fiselier <eric at efcs.ca> wrote:
>>>>>
>>>>>> +rsmith (actually this time)
>>>>>>
>>>>>> On Sun, Sep 30, 2018 at 12:09 PM Eric Fiselier <eric at efcs.ca> wrote:
>>>>>>
>>>>>>> +rsmith
>>>>>>>
>>>>>>> Hi All,
>>>>>>>
>>>>>>> Sorry, I'm not actually sure why this fix is correct.I stole both
>>>>>>> the fix and the comment from a similar one on L150 of the module map left
>>>>>>> by Richard Smith.
>>>>>>>
>>>>>>> /Eric
>>>>>>>
>>>>>>> On Tue, Sep 25, 2018 at 8:36 PM Shuai Wang <shuaiwang at google.com>
>>>>>>> wrote:
>>>>>>>
>>>>>>>> I'd like to understand this better as well, in particular what
>>>>>>>> would be a proper fix?
>>>>>>>>
>>>>>>>> On Tue, Sep 25, 2018 at 2:15 PM David Blaikie <dblaikie at gmail.com>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>>> +Shuai Wang
>>>>>>>>>
>>>>>>>>> On Tue, Sep 25, 2018 at 2:14 PM David Blaikie <dblaikie at gmail.com>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>> Hey Eric - thanks for the fix - but could you explain the issue
>>>>>>>>>> here in a bit more detail, as I'm a bit confused (& really interested in
>>>>>>>>>> understanding any layering problems in LLVM - and fixing them/making sure
>>>>>>>>>> they're fixed/holding the line/etc)
>>>>>>>>>>
>>>>>>>>>> What do you mean by "pull all of the AST matchers library into
>>>>>>>>>> clang" - how does including a header ever add a link dependency?
>>>>>>>>>>
>>>>>>>>>> - Dave
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On Sat, Sep 22, 2018 at 5:49 PM Eric Fiselier via cfe-commits <
>>>>>>>>>> cfe-commits at lists.llvm.org> wrote:
>>>>>>>>>>
>>>>>>>>>>> Author: ericwf
>>>>>>>>>>> Date: Sat Sep 22 17:48:05 2018
>>>>>>>>>>> New Revision: 342827
>>>>>>>>>>>
>>>>>>>>>>> URL: http://llvm.org/viewvc/llvm-project?rev=342827&view=rev
>>>>>>>>>>> Log:
>>>>>>>>>>> Fix modules build with shared library.
>>>>>>>>>>>
>>>>>>>>>>> r341994 caused clangAnalysis to pull all of the AST matchers
>>>>>>>>>>> library into clang. Due to inline key functions in the headers,
>>>>>>>>>>> importing the AST matchers library gives a link dependency on the
>>>>>>>>>>> AST matchers (and thus the AST), which clang should not
>>>>>>>>>>> have.
>>>>>>>>>>>
>>>>>>>>>>> This patch works around the issues by excluding the offending
>>>>>>>>>>> libclangAnalysis header in the modulemap.
>>>>>>>>>>>
>>>>>>>>>>> Modified:
>>>>>>>>>>>     cfe/trunk/include/clang/module.modulemap
>>>>>>>>>>>
>>>>>>>>>>> Modified: cfe/trunk/include/clang/module.modulemap
>>>>>>>>>>> URL:
>>>>>>>>>>> http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/module.modulemap?rev=342827&r1=342826&r2=342827&view=diff
>>>>>>>>>>>
>>>>>>>>>>> ==============================================================================
>>>>>>>>>>> --- cfe/trunk/include/clang/module.modulemap (original)
>>>>>>>>>>> +++ cfe/trunk/include/clang/module.modulemap Sat Sep 22 17:48:05
>>>>>>>>>>> 2018
>>>>>>>>>>> @@ -5,6 +5,12 @@ module Clang_Analysis {
>>>>>>>>>>>    textual header "Analysis/Analyses/ThreadSafetyOps.def"
>>>>>>>>>>>
>>>>>>>>>>>    module * { export * }
>>>>>>>>>>> +
>>>>>>>>>>> +  // FIXME: Exclude these headers to avoid pulling all of the
>>>>>>>>>>> AST matchers
>>>>>>>>>>> +  // library into clang. Due to inline key functions in the
>>>>>>>>>>> headers,
>>>>>>>>>>> +  // importing the AST matchers library gives a link dependency
>>>>>>>>>>> on the AST
>>>>>>>>>>> +  // matchers (and thus the AST), which clang-format should not
>>>>>>>>>>> have.
>>>>>>>>>>> +  exclude header "Analysis/Analyses/ExprMutationAnalyzer.h"
>>>>>>>>>>>  }
>>>>>>>>>>>
>>>>>>>>>>>  module Clang_AST {
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> cfe-commits mailing list
>>>>>>>>>>> cfe-commits at lists.llvm.org
>>>>>>>>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>>>>>>>>>>>
>>>>>>>>>> _______________________________________________
>>>>> cfe-commits mailing list
>>>>> cfe-commits at lists.llvm.org
>>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>>>>>
>>>> _______________________________________________
>>>> cfe-commits mailing list
>>>> cfe-commits at lists.llvm.org
>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>>>>
>>>>
>>>> _______________________________________________
>>>> cfe-commits mailing list
>>>> cfe-commits at lists.llvm.org
>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>>>>
>>> _______________________________________________
>> cfe-commits mailing list
>> cfe-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>>
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20181022/edd44385/attachment-0001.html>

More information about the cfe-commits mailing list