<div dir="ltr">If you want this in the core libs, a great idea would be to write up a doc or email explaining the design and the use cases.<div>Note that if this doesn't solve a generic enough use case, it might still be a good idea for it to live in a different place - API surface is a cost to users.</div></div><br><div class="gmail_quote"><div dir="ltr">On Tue, Oct 23, 2018 at 5:56 PM Jonas Toth via cfe-dev <<a href="mailto:cfe-dev@lists.llvm.org">cfe-dev@lists.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div text="#000000" bgcolor="#FFFFFF">
<p>Hi Rafael,</p>
<p>nice to see your progress!</p>
<p>In my opinion these advancements should go in the Tooling and
Rewrite directly in clang but integrate with the existing
facilities. It would be inconvenient two have two separate ways of
doing rewritings.</p>
<p>What I am curious about is how to resolve the macro-situation. If
the refactoring will work directly on the AST and ignore macros,
how is the refactoring translated into real code again?</p>
<p>Having an functional way of doing the refactorings with just
transforming the AST, e.g. replacing one node with a new subtree
would be great and elegant on the AST side, but how would we move
back from AST to code?</p>
<p>Best, Jonas<br>
</p>
<div class="m_-273721086631250889moz-cite-prefix">Am 23.10.18 um 17:32 schrieb
Rafael·Stahl:<br>
</div>
<blockquote type="cite">
<p>Hey everyone</p>
<p>the two major limitations are resolved now:</p>
<p>- The macro argument issue<br>
- Support for replacing Decls using
clang::ast_type_traits::DynTypedNode<br>
</p>
<p>The macro issue vanished by itself once I figured a little
unintuitive detail out: The SourceLocations from the original
AST and the ones from the new Preprocessor Lexer did not compare
equal even though they refer to the exact same location.</p>
<p>I believe this is because expansion SrcLocs have a kind of
pointer identity representation and equality just compares raw
representations, such that even when they point to the same
locations, they don't compare equal. What worked for me was a
kind of "deep" comparison:<br>
</p>
<p>bool clutil::DeepSrcLocEqual(clang::SourceLocation lhs,
clang::SourceLocation rhs, const clang::SourceManager &SM)<br>
{<br>
if (lhs == rhs)<br>
return true;<br>
<br>
if (SM.getExpansionLoc(lhs) != SM.getExpansionLoc(rhs))<br>
return false;<br>
if (SM.getSpellingLoc(lhs) != SM.getSpellingLoc(rhs))<br>
return false;<br>
<br>
clang::SourceLocation lhsMacro, rhsMacro;<br>
if (SM.isMacroArgExpansion(lhs, &lhsMacro))<br>
{<br>
if (!SM.isMacroArgExpansion(rhs, &rhsMacro))<br>
return false;<br>
if (!DeepSrcLocEqual(lhsMacro, rhsMacro, SM))<br>
return false;<br>
}<br>
<br>
return true;<br>
}</p>
<p>Attached is the cleaned up rewriter that I ended up with now.
Still, if there is any interest, I'd be happy to contribute this
back to clangs libraries, but I would require some feedback how
this fits into existing facilities.<br>
</p>
<p>Best regards<br>
Rafael<br>
</p>
<div class="m_-273721086631250889moz-cite-prefix">On 17.07.18 16:19, Jonas Toth wrote:<br>
</div>
<blockquote type="cite">
<p>Hi Rafael,</p>
<p>I did read into clang-refactor a while ago but unfortunatly
could not follow that up. If I recall correctly its about
source-to-source transformation (as you said) and aims at
implementing the primitive refactorings that exist (e.g.
extract-method, extract-variable, ....).</p>
<p>Rewriting itself should happen with the normal tooling
framework.<br>
</p>
(<a class="m_-273721086631250889moz-txt-link-freetext" href="https://clang.llvm.org/docs/RefactoringEngine.html" target="_blank">https://clang.llvm.org/docs/RefactoringEngine.html</a>)<br>
<br>
Maybe the implementers of the existing code can give better
comments on you proposal (and might have considered a similar
solution to yours already). <br>
<br>
+Alex Lorenz<br>
<br>
All the best, Jonas<br>
<br>
<br>
<div class="m_-273721086631250889moz-cite-prefix">Am 17.07.2018 um 14:46 schrieb
Rafael·Stahl:<br>
</div>
<blockquote type="cite">
<p>Hi Jonas</p>
<p>Thanks for introducing me to this, I have seen the
"Replacement" before, but not clang-refactor.</p>
<p>However it seems to only provide management facilities
around rewrite operations and not aid with the rewriting
itself. Am I missing something here?</p>
<p>The two core problems for me:</p>
<p>- nesting replacements: When implementing replacements with
clang-refactor, I still have to provide replacements that
are closed in themselves. I cannot make them depend on
others, right?<br>
- macros: clang-refactor only seems to work with spelling
locations.</p>
<p>Maybe an even simpler example: Replace all additions with
"add(lhs, rhs)". This in itself is very difficult with clang
as soon as the Stmts are nested or macros are involved.<br>
</p>
<p>Best regards<br>
Rafael<br>
</p>
<br>
<div class="m_-273721086631250889moz-cite-prefix">On 16.07.2018 19:06, Jonas Toth
via cfe-dev wrote:<br>
</div>
<blockquote type="cite">
<p>Hi Rafael,</p>
<p>wouldn't your usecase be a task for clang-refactor?</p>
<p>Best, Jonas<br>
</p>
<br>
<div class="m_-273721086631250889moz-cite-prefix">Am 16.07.2018 um 17:08 schrieb
Rafael·Stahl via cfe-dev:<br>
</div>
<blockquote type="cite">Hey
everyone <br>
<br>
The rewriting API of Clang operates on the source code in
textual form. The user can use AST nodes to figure out
what to replace, but in the end he has to remove and
insert snippets in a linear piece of text. <br>
<br>
This is very inconvenient when it is required to
restructure and nest replacements. The involvement of
macros makes a manual process even more difficult. See
some recent threads expressing difficulty with the API
[1][2]. <br>
<br>
What do I mean by "nested replacements"? For example in
the following: <br>
<br>
int i = x + s->a; <br>
<br>
I would want to replace the BinaryOperator with a function
call and the MemberExpr with some constant: <br>
<br>
int i = Addition(x, 7); <br>
<br>
When keeping the two replacement rules independent of each
other, achieving this with the current API is extremely
difficult. More so when macros are involved. <br>
<br>
I am proposing some kind of helper that aims to solve
these issues by providing an interface that offers to
directly replace AST nodes and a mechanism to nest AST
node replacements - without having to worry about macros.
<br>
<br>
Potential usage: <br>
<br>
- Develop a class that derives from StmtToRewrite to
define how replacements should happen: <br>
<br>
class RewriteAdds : public cu::StmtToRewrite <br>
{ <br>
public: <br>
std::string makeReplaceStr() const override <br>
{ <br>
auto binOp =
dyn_cast<BinaryOperator>(replaceS); <br>
return "Addition(" +
getMgr()->getReplaced(binOp->getLHS()).strToInsert +
", " + <br>
getMgr()->getReplaced(binOp->getRHS()).strToInsert +
")"; <br>
} <br>
}; <br>
<br>
class RewriteMembs : public cu::StmtToRewrite <br>
{ <br>
public: <br>
std::string makeReplaceStr() const override <br>
{ <br>
return "7"; <br>
} <br>
}; <br>
<br>
- Construct a RewriteManager: <br>
<br>
cu::RewriteManager mgr(ACtx, PP); <br>
<br>
- Add rewriting operations to the manager: <br>
<br>
// std::vector<const Stmt *> AddStmts = /*
matched from binaryOperator() with plus */ <br>
// std::vector<const Stmt *> MembStmts = /*
matched from memberExpr() */ <br>
for (const auto &S : AddStmts)
mgr.registerStmt<RewriteAdds>(S); <br>
for (const auto &S : MembStmts)
mgr.registerStmt<RewriteMembs>(S); <br>
<br>
- Retrieve and apply the results: <br>
<br>
clang::Rewriter rewriter(SM, LangOpts); <br>
for (const auto &r : mgr.getReplacements()) { <br>
rewriter.RemoveText(r.rangeToRemove); <br>
rewriter.InsertText(r.rangeToRemove.getBegin(),
r.strToInsert); <br>
} <br>
<br>
<br>
At the end of this mail is my low quality code that
kind-of implements this. TLDR: <br>
<br>
- Build a hierarchy of stmts to replace and keep track of
which replacements must be combined <br>
- Move further up in the AST if these replacements are
inside a macro <br>
- Recursively lex the file and look for replacements
outside-in by spelling locations. Expand any macros that
are encountered during this. The re-lexing idea is based
on the hint in [3]. <br>
<br>
The code has the following shortcomings: <br>
<br>
- I do not know how to distinguish macro argument
expansions within macros. For example in "#define FOO(a) a
+ a" the two "a"s expand to different AST nodes that could
be replaced with different rules. This is an important
issue, because it can lead to completely broken code with
nesting. <br>
- Limited to Stmts, when Decls should be supported too. <br>
- Very un-optimized with lexing the entire source file
many times. Easy to solve, but didn't want to raise the
complexity further for now. <br>
- Could keep written code more clean by only expanding
macros if required. For example not required if just a
macro arg is replaced and all expansions would be the
same. <br>
<br>
<br>
I am very interested in your general thoughts. I'm not
very experienced with clang, but this was my vision how I
would want to do replacements. Are you interested in
getting this into clang? I would need help with ironing
out the remaining issues. <br>
<br>
-Rafael <br>
<br>
<br>
[1] <a class="m_-273721086631250889moz-txt-link-freetext" href="http://lists.llvm.org/pipermail/cfe-dev/2018-July/058430.html" target="_blank">http://lists.llvm.org/pipermail/cfe-dev/2018-July/058430.html</a>
<br>
[2] <a class="m_-273721086631250889moz-txt-link-freetext" href="http://lists.llvm.org/pipermail/cfe-dev/2018-June/058213.html" target="_blank">http://lists.llvm.org/pipermail/cfe-dev/2018-June/058213.html</a>
<br>
[3] <a class="m_-273721086631250889moz-txt-link-freetext" href="http://lists.llvm.org/pipermail/cfe-dev/2017-August/055079.html" target="_blank">http://lists.llvm.org/pipermail/cfe-dev/2017-August/055079.html</a>
<br>
<br>
<br>
<br>
---------------------------------------- <br>
<br>
RewriteManager.h <br>
<br>
---------------------------------------- <br>
<br>
#ifndef CLANGUTIL_REWRITEMANAGER_H <br>
#define CLANGUTIL_REWRITEMANAGER_H <br>
<br>
#include "ClangUtil/SourceRangeLess.h" <br>
#include "make_unique.h" <br>
#include "clang/AST/AST.h" <br>
#include <vector> <br>
#include <map> <br>
<br>
<br>
// TODO extend to decls <br>
<br>
<br>
namespace cu <br>
{ <br>
// Represents a statement in the original AST that should
be rewritten. To implement recursive replacements, call <br>
// getMgr()->getReplaced() on any AST node within the
makeReplaceStr callback. <br>
class StmtToRewrite <br>
{ <br>
friend class RewriteManager; <br>
<br>
public: <br>
// Returns the enclosing RewriteManager. <br>
class RewriteManager *getMgr() const; <br>
// Override this to build a replacement string.
Implement recursive replacements with
RewriteManager::getReplaced. <br>
virtual std::string makeReplaceStr() const = 0; <br>
<br>
// The statement to replace. <br>
const clang::Stmt *replaceS = nullptr; <br>
<br>
private: <br>
RewriteManager *m_mgr; <br>
}; <br>
<br>
struct RewriteOperation <br>
{ <br>
clang::SourceRange rangeToRemove; <br>
std::string strToInsert; <br>
}; <br>
<br>
// A class for managing replacements of AST nodes. It
allows to specifically target AST nodes instead of raw
source <br>
// locations to enable easy replacements involving macros
and nested replacements. <br>
// For extended documentation see: doc/rewriting.md <br>
class RewriteManager <br>
{ <br>
public: <br>
RewriteManager(clang::ASTContext &ACtx,
clang::Preprocessor &PP); <br>
<br>
clang::ASTContext &getACtx() const { return ACtx;
} <br>
<br>
// Registers a StmtToRewrite for use with
getReplacements. Call this on all <br>
// statements that should be rewritten before calling
any rewriting functions. <br>
void registerStmt(std::unique_ptr<StmtToRewrite>
S); <br>
<br>
// Helper for constructing the custom type from a
Stmt. <br>
template <typename T, typename... Args> <br>
void registerStmt(const clang::Stmt *S, Args... args)
<br>
{ <br>
auto p =
std::make_unique<T>(std::forward<Args>(args)...);
<br>
p->replaceS = S; <br>
registerStmt(std::move(p)); <br>
} <br>
<br>
// Get the full replacement of an AST node. Note that
this function removes any replaced statements from the
work <br>
// list, so calling it twice will only replace the
first time. <br>
RewriteOperation getReplaced(const clang::Stmt *S); <br>
// Get all replacements. These may be fewer than the
requested ones because of nesting. <br>
std::vector<RewriteOperation> getReplacements();
<br>
<br>
private: <br>
std::string getExpandedCode(const clang::Stmt
*toReplaceS); <br>
<br>
private: <br>
clang::ASTContext &ACtx; <br>
const clang::LangOptions &LangOpts; <br>
clang::SourceManager &SM; <br>
clang::Preprocessor &PP; <br>
<br>
// Manages the pending replacements. <br>
class WorkList <br>
{ <br>
public: <br>
typedef std::map<clang::SourceRange,
std::vector<const StmtToRewrite *>>
RangeToRepMap; <br>
<br>
WorkList(clang::ASTContext &ACtx,
clang::SourceManager &SM); <br>
<br>
bool isStmtPending(const clang::Stmt *S) const; <br>
void addStmt(std::unique_ptr<StmtToRewrite>
S); <br>
const RangeToRepMap
&getRangeToReplacementsMap() const; <br>
std::vector<const StmtToRewrite *>
getSortedReplacements() const; <br>
void markDone(const StmtToRewrite *S); <br>
void cleanup(); <br>
<br>
private: <br>
clang::ASTContext &ACtx; <br>
clang::SourceManager &SM; <br>
std::vector<std::unique_ptr<StmtToRewrite>>
m_pending; <br>
std::vector<std::unique_ptr<StmtToRewrite>>
m_done; <br>
RangeToRepMap m_rangeToReplacements; <br>
}; <br>
<br>
WorkList m_workList; <br>
}; <br>
<br>
} // namespace cu <br>
<br>
#endif <br>
<br>
<br>
<br>
---------------------------------------- <br>
<br>
RewriteManager.cpp <br>
<br>
---------------------------------------- <br>
<br>
#include "ClangUtil/RewriteManager.h" <br>
#include "ClangUtil/ASTUtil.h" <br>
#include "clang/Lex/Lexer.h" <br>
#include "clang/Lex/Preprocessor.h" <br>
#include "clang/Lex/PreprocessorOptions.h" <br>
#include "clang/Lex/TokenConcatenation.h" <br>
#include "clang/Lex/MacroArgs.h" <br>
<br>
<br>
using namespace cu; <br>
<br>
<br>
// Returns a Stmt that is the first parent of startS whose
expansion range is within the given range. <br>
static const clang::Stmt
*GetFullMacroStmt(clang::SourceRange range, const
clang::Stmt *startS, clang::ASTContext &ACtx) <br>
{ <br>
auto &SM = ACtx.getSourceManager(); <br>
<br>
// Walk the tree upwards until ST does no longer
expand to within range. <br>
const clang::Stmt *ST = startS; <br>
while (true) <br>
{ <br>
const auto &parents = ACtx.getParents(*ST); <br>
if (parents.empty()) <br>
{ <br>
break; <br>
} <br>
auto childS = ST; <br>
ST = parents[0].get<clang::Stmt>(); <br>
if (!ST) <br>
{ <br>
if (auto D =
parents[0].get<clang::Decl>()) <br>
{ <br>
const auto &parentsD =
ACtx.getParents(*D); <br>
if (parentsD.empty()) <br>
{ <br>
break; <br>
} <br>
ST = parentsD[0].get<clang::Stmt>();
<br>
if (!ST) <br>
{ <br>
break; <br>
} <br>
} <br>
else <br>
{ <br>
break; <br>
} <br>
} <br>
<br>
auto exLocS =
SM.getExpansionLoc(ST->getLocStart()); <br>
auto exLocE =
SM.getExpansionLoc(ST->getLocEnd()); <br>
if (SM.isBeforeInTranslationUnit(exLocS,
range.getBegin()) || <br>
SM.isBeforeInTranslationUnit(range.getEnd(),
exLocE)) <br>
{ <br>
return childS; <br>
} <br>
} <br>
<br>
return nullptr; <br>
} <br>
<br>
<br>
RewriteManager *StmtToRewrite::getMgr() const <br>
{ <br>
return m_mgr; <br>
} <br>
<br>
<br>
RewriteManager::WorkList::WorkList(clang::ASTContext
&ACtx, clang::SourceManager &SM) : ACtx(ACtx),
SM(SM) {} <br>
bool RewriteManager::WorkList::isStmtPending(const
clang::Stmt *S) const <br>
{ <br>
for (const auto &r : m_pending) <br>
{ <br>
if (r->replaceS == S) <br>
{ <br>
return true; <br>
} <br>
} <br>
return false; <br>
} <br>
void
RewriteManager::WorkList::addStmt(std::unique_ptr<StmtToRewrite>
S) <br>
{ <br>
// Use the expansion range for maximal replacement
flexibility in macros. <br>
auto replaceRange =
SM.getExpansionRange(S->replaceS->getSourceRange());
<br>
<br>
// TODO not quite correct. <br>
/*auto sortRanges = [&](std::vector<const
StmtToRewrite *> &vec) { <br>
std::sort(vec.begin(), vec.end(), [&](const
StmtToRewrite *lhs, const StmtToRewrite *rhs) { <br>
auto lhsRange =
SM.getExpansionRange(lhs->replaceS->getSourceRange());
<br>
auto rhsRange =
SM.getExpansionRange(rhs->replaceS->getSourceRange());
<br>
return IsContained(rhsRange, lhsRange, SM); <br>
}); <br>
};*/ <br>
<br>
// Establish hierarchical relation between all ranges.
<br>
bool found = false; <br>
// First, check if this range is within one we already
have. <br>
for (auto &r : m_rangeToReplacements) <br>
{ <br>
if (IsContained(replaceRange, r.first, SM)) <br>
{ <br>
// Insert in a sorted order. <br>
for (auto it = r.second.begin(); it !=
r.second.end(); ++it) <br>
{ <br>
//auto testRange =
SM.getExpansionRange((*it)->replaceS->getSourceRange());
<br>
// if (IsContained(testRange,
replaceRange, SM)) <br>
if (IsParent(S->replaceS,
(*it)->replaceS, ACtx)) <br>
{ <br>
r.second.insert(it, S.get()); <br>
found = true; <br>
break; <br>
} <br>
} <br>
if (!found) <br>
{ <br>
r.second.push_back(S.get()); <br>
found = true; <br>
} <br>
break; <br>
} <br>
} <br>
// Not within existing range, add as new top-level
range. <br>
if (!found) <br>
{ <br>
// Check if any existing ranges are contained
within the new one. <br>
std::vector<const StmtToRewrite *>
moveThese; <br>
auto it = m_rangeToReplacements.begin(); <br>
while (it != m_rangeToReplacements.end()) <br>
{ <br>
if (IsContained(it->first, replaceRange,
SM)) <br>
{ <br>
moveThese.insert(moveThese.end(),
it->second.begin(), it->second.end()); <br>
it = m_rangeToReplacements.erase(it); <br>
} <br>
else <br>
{ <br>
++it; <br>
} <br>
} <br>
auto &accesses =
m_rangeToReplacements[replaceRange]; <br>
// The order is important here. We want the first
element to be the one that spans the full range. <br>
accesses.push_back(S.get()); <br>
// TODO sort "moveThese". <br>
accesses.insert(accesses.end(), moveThese.begin(),
moveThese.end()); <br>
} <br>
<br>
int count = 0; <br>
for (const auto &r : m_rangeToReplacements) <br>
{ <br>
printf("range %i\n", count++); <br>
for (const auto &a : r.second) <br>
{ <br>
printf("replacement:\n"); <br>
a->replaceS->dump(); <br>
} <br>
} <br>
<br>
m_pending.push_back(std::move(S)); <br>
} <br>
const RewriteManager::WorkList::RangeToRepMap
&RewriteManager::WorkList::getRangeToReplacementsMap()
const <br>
{ <br>
return m_rangeToReplacements; <br>
} <br>
std::vector<const StmtToRewrite *>
RewriteManager::WorkList::getSortedReplacements() const <br>
{ <br>
std::vector<const StmtToRewrite *> result; <br>
for (auto &r : m_rangeToReplacements) <br>
{ <br>
result.insert(result.end(), r.second.begin(),
r.second.end()); <br>
} <br>
return result; <br>
} <br>
void RewriteManager::WorkList::markDone(const
StmtToRewrite *S) <br>
{ <br>
// Remove from hierarchy. <br>
for (auto &r : m_rangeToReplacements) <br>
{ <br>
r.second.erase(std::remove(r.second.begin(),
r.second.end(), S), r.second.end()); <br>
} <br>
<br>
// Move from pending to done list. <br>
auto it = std::find_if(m_pending.begin(),
m_pending.end(), <br>
[&](const
std::unique_ptr<StmtToRewrite> &rep) { return
rep.get() == S; }); <br>
if (it == m_pending.end()) <br>
{ <br>
throw std::runtime_error("Did not find replacement
to mark as done"); <br>
} <br>
m_done.push_back(std::move(*it)); <br>
m_pending.erase(it); <br>
} <br>
void RewriteManager::WorkList::cleanup() <br>
{ <br>
m_done.clear(); <br>
} <br>
<br>
<br>
RewriteManager::RewriteManager(clang::ASTContext
&ACtx, clang::Preprocessor &PP) <br>
: ACtx(ACtx), LangOpts(ACtx.getLangOpts()),
SM(ACtx.getSourceManager()), PP(PP), m_workList(ACtx, SM)
<br>
{ <br>
} <br>
<br>
void
RewriteManager::registerStmt(std::unique_ptr<StmtToRewrite>
S) <br>
{ <br>
if (!S->replaceS) <br>
{ <br>
throw std::runtime_error("Must set replaceS"); <br>
} <br>
<br>
if (m_workList.isStmtPending(S->replaceS)) <br>
{ <br>
throw std::runtime_error("This Stmt will already
be replaced"); <br>
} <br>
<br>
S->m_mgr = this; <br>
m_workList.addStmt(std::move(S)); <br>
} <br>
<br>
RewriteOperation RewriteManager::getReplaced(const
clang::Stmt *S) <br>
{ <br>
auto range =
SM.getExpansionRange(S->getSourceRange()); <br>
return { range, getExpandedCode(S) }; <br>
} <br>
<br>
std::vector<RewriteOperation>
RewriteManager::getReplacements() <br>
{ <br>
std::vector<RewriteOperation> results; <br>
<br>
for (auto &rangeAndAccesses :
m_workList.getRangeToReplacementsMap()) <br>
{ <br>
auto &range = rangeAndAccesses.first; <br>
auto &accesses = rangeAndAccesses.second; <br>
<br>
// Cannot replace something inside a macro because
it would replace all expansions instead of just the
selected <br>
// AST node. So in a first step, get an enclosing
statement that is no longer inside a macro. <br>
// TODO we could keep the original code more clean
by not expanding macro args if the whole expansion does
not <br>
// contain the macro arg more than once. <br>
auto macroS = GetFullMacroStmt(range,
accesses[0]->replaceS, ACtx); <br>
<br>
results.push_back(getReplaced(macroS)); <br>
<br>
// TODO we could run clang-format on the
replacements. this would especially benefit long macro
expansions. <br>
} <br>
<br>
m_workList.cleanup(); <br>
<br>
return results; <br>
} <br>
<br>
std::string RewriteManager::getExpandedCode(const
clang::Stmt *toReplaceS) <br>
{ <br>
// TODO performance optimization. this is parsing way
more than required. <br>
<br>
using namespace clang; <br>
<br>
printf("getExpandedCode:\n"); <br>
toReplaceS->dump(); <br>
<br>
std::string out; <br>
<br>
auto toReplaceExpStart =
SM.getExpansionLoc(toReplaceS->getLocStart()); <br>
auto toReplaceExpEnd =
SM.getExpansionLoc(toReplaceS->getLocEnd()); <br>
auto toReplaceSpellStart =
SM.getSpellingLoc(toReplaceS->getLocStart()); <br>
auto toReplaceSpellEnd =
SM.getSpellingLoc(toReplaceS->getLocEnd()); <br>
<br>
auto FID =
SM.getFileID(SM.getExpansionLoc(toReplaceS->getLocStart()));
<br>
<br>
// The following is inspired by:
clang/Rewrite/HTMLRewrite.cpp:HighlightMacros <br>
<br>
// Re-lex the raw token stream into a token buffer. <br>
std::vector<Token> TokenStream; <br>
<br>
const llvm::MemoryBuffer *FromFile =
SM.getBuffer(FID); <br>
Lexer L(FID, FromFile, SM, PP.getLangOpts()); <br>
<br>
// Lex all the tokens in raw mode, to avoid entering
#includes or expanding <br>
// macros. <br>
while (1) <br>
{ <br>
Token Tok; <br>
L.LexFromRawLexer(Tok); <br>
<br>
// If this is a # at the start of a line, discard
it from the token stream. <br>
// We don't want the re-preprocess step to see
#defines, #includes or other <br>
// preprocessor directives. <br>
if (Tok.is(tok::hash) &&
Tok.isAtStartOfLine()) <br>
continue; <br>
<br>
// If this is a ## token, change its kind to
unknown so that repreprocessing <br>
// it will not produce an error. <br>
if (Tok.is(tok::hashhash)) <br>
Tok.setKind(tok::unknown); <br>
<br>
// If this raw token is an identifier, the raw
lexer won't have looked up <br>
// the corresponding identifier info for it. Do
this now so that it will be <br>
// macro expanded when we re-preprocess it. <br>
if (Tok.is(tok::raw_identifier)) <br>
PP.LookUpIdentifierInfo(Tok); <br>
<br>
TokenStream.push_back(Tok); <br>
<br>
for (auto &rep :
m_workList.getSortedReplacements()) <br>
{ <br>
auto repS = rep->replaceS; <br>
auto spellLoc =
SM.getSpellingLoc(repS->getLocStart()); <br>
if (SM.getSpellingLoc(Tok.getLocation()) ==
spellLoc) <br>
{ <br>
// <br>
} <br>
} <br>
<br>
if (Tok.is(tok::eof)) <br>
break; <br>
} <br>
<br>
// Temporarily change the diagnostics object so that
we ignore any generated <br>
// diagnostics from this pass. <br>
DiagnosticsEngine
TmpDiags(PP.getDiagnostics().getDiagnosticIDs(),
&PP.getDiagnostics().getDiagnosticOptions(), <br>
new IgnoringDiagConsumer);
<br>
<br>
// Copy the preprocessor and all of its state. <br>
auto PPOpts =
std::make_shared<PreprocessorOptions>(PP.getPreprocessorOpts());
<br>
LangOptions LO = PP.getLangOpts(); <br>
Preprocessor TmpPP(PPOpts, TmpDiags, LO, SM,
PP.getPCMCache(), PP.getHeaderSearchInfo(),
PP.getModuleLoader(), <br>
PP.getIdentifierTable().getExternalIdentifierLookup()); <br>
TmpPP.Initialize(PP.getTargetInfo(),
PP.getAuxTargetInfo()); <br>
TmpPP.setExternalSource(PP.getExternalSource()); <br>
TmpPP.setPreprocessedOutput(true); <br>
<br>
std::map<const clang::IdentifierInfo *, bool>
MacroPreviouslyEnabled; <br>
for (const auto &m : PP.macros()) <br>
{ <br>
// printf("PREDEF MACRO: %s\n",
m.first->getName().str().c_str()); <br>
TmpPP.getMacroDefinition(m.first); <br>
<br>
for (const auto &tmpm : TmpPP.macros()) <br>
{ <br>
if (tmpm.first == m.first) <br>
{ <br>
auto MD = m.second.getLatest(); <br>
auto MI = MD->getMacroInfo(); <br>
// If this is a recursive call we might be
in a macro expansion and the macro might be disabled. We
need <br>
// to enable it for now so that all
expansions work. Restore it later. <br>
MacroPreviouslyEnabled[tmpm.first] =
MI->isEnabled(); <br>
if (!MI->isEnabled()) <br>
{ <br>
MD->getMacroInfo()->EnableMacro(); <br>
} <br>
<br>
// This should not change anything since
we just copy data over. <br>
auto &mutableState =
const_cast<std::remove_const<decltype(tmpm.second)>::type
&>(tmpm.second); <br>
mutableState.setLatest(MD); <br>
break; <br>
} <br>
} <br>
} <br>
<br>
class MacroArgCollector : public clang::PPCallbacks <br>
{ <br>
public: <br>
MacroArgCollector(Preprocessor &TmpPP) :
TmpPP(TmpPP) {} <br>
<br>
void MacroExpands(const Token &Tok, const
MacroDefinition &MD, SourceRange Range, const
MacroArgs *Args) override <br>
{ <br>
if (!Args) <br>
{ <br>
return; <br>
} <br>
printf("GOT MACRO ARGS EXPANSION CALLBACK\n");
<br>
for (int i = 0; i <
(int)Args->getNumMacroArguments(); i++) <br>
{ <br>
auto TokUnex =
Args->getUnexpArgument(i); <br>
// Thats just non-const for a cache, so
should be fine. <br>
auto TokPreExp = const_cast<MacroArgs
*>(Args)->getPreExpArgument(i, TmpPP); <br>
printf("unexp: %s\n",
TmpPP.getSpelling(*TokUnex).c_str()); <br>
for (const auto &T : TokPreExp) <br>
{ <br>
printf("preexp: %s\n",
TmpPP.getSpelling(T).c_str()); <br>
} <br>
} <br>
} <br>
<br>
Preprocessor &TmpPP; <br>
}; <br>
TmpPP.addPPCallbacks(std::make_unique<MacroArgCollector>(TmpPP));
<br>
// Instead: collect the macro arg info in the law
lexing step above. or do another pass that uses the PP but
without expansions. <br>
<br>
/*printf("DUMP MACRO INFO\n"); <br>
for (const auto &m : PP.macros()) <br>
PP.dumpMacroInfo(m.first); <br>
printf("---\n"); <br>
for (const auto &m : TmpPP.macros()) <br>
TmpPP.dumpMacroInfo(m.first); <br>
printf("DUMP MACRO INFO END\n");*/ <br>
<br>
DiagnosticsEngine *OldDiags =
&TmpPP.getDiagnostics(); <br>
<br>
// Inform the preprocessor that we don't want
comments. <br>
TmpPP.SetCommentRetentionState(false, false); <br>
<br>
// We don't want pragmas either. Although we filtered
out #pragma, removing <br>
// _Pragma and __pragma is much harder. <br>
bool PragmasPreviouslyEnabled =
TmpPP.getPragmasEnabled(); <br>
TmpPP.setPragmasEnabled(false); <br>
<br>
// Enter the tokens we just lexed. This will cause
them to be macro expanded <br>
// but won't enter sub-files (because we removed #'s).
<br>
TmpPP.EnterTokenStream(TokenStream, false); <br>
<br>
TokenConcatenation ConcatInfo(TmpPP); <br>
<br>
// Lex all the tokens. <br>
Token Tok; <br>
TmpPP.Lex(Tok); <br>
<br>
std::map<SourceLocation, int> slocIdx; <br>
<br>
auto checkReplacement = [&]() { <br>
for (auto &rep :
m_workList.getSortedReplacements()) <br>
{ <br>
// auto rep = r.second.get(); <br>
auto repS = rep->replaceS; <br>
auto spellLoc =
SM.getSpellingLoc(repS->getLocStart()); <br>
// TODO we need to check here if the repS
spans the full range (or largest?) <br>
if (SM.getSpellingLoc(Tok.getLocation()) ==
spellLoc) <br>
{ <br>
if (slocIdx[spellLoc] == 7) <br>
{ <br>
// replace <br>
} <br>
slocIdx[spellLoc]++; <br>
<br>
// Done replacing that one, but have to
keep it alive until we're done with it. <br>
m_workList.markDone(rep); <br>
<br>
printf("[[[\n"); <br>
auto repStr = rep->makeReplaceStr(); <br>
printf("REPLACED: %s ]]]\n",
repStr.c_str()); <br>
out += repStr; <br>
<br>
// Skip ahead until after the whole
replacement. <br>
auto repEnd =
SM.getSpellingLoc(repS->getLocEnd()); <br>
while (repEnd !=
SM.getSpellingLoc(Tok.getLocation())) <br>
{ <br>
TmpPP.Lex(Tok); <br>
assert(!Tok.is(tok::eof) &&
"End not found"); <br>
} <br>
<br>
// Eat one more since we stopped at the
end token and we want to continue after it. <br>
TmpPP.Lex(Tok); <br>
<br>
return true; <br>
} <br>
} <br>
return false; <br>
}; <br>
<br>
while (Tok.isNot(tok::eof)) <br>
{ <br>
printf("TOKEN: %s\n",
TmpPP.getSpelling(Tok).c_str()); <br>
<br>
auto TokLoc = Tok.getLocation(); <br>
auto TokExp = SM.getExpansionLoc(TokLoc); <br>
if (SM.isBeforeInTranslationUnit(toReplaceExpEnd,
TokExp)) <br>
{ <br>
// Anything after the Stmt we want to replace
is not interesting. <br>
break; <br>
} <br>
<br>
// Skip ahead until we are at the expansion start
of the Stmt we want to replace. <br>
if (!SM.isBeforeInTranslationUnit(TokLoc,
toReplaceExpStart)) <br>
{ <br>
if (TokLoc.isMacroID()) <br>
{ <br>
// This is the first token of a macro
expansion. <br>
auto LLoc = SM.getExpansionRange(TokLoc);
<br>
<br>
// Ignore tokens whose instantiation
location was not the main file. <br>
if (SM.getFileID(LLoc.first) != FID) <br>
{ <br>
TmpPP.Lex(Tok); <br>
continue; <br>
} <br>
<br>
assert(SM.getFileID(LLoc.second) == FID
&& <br>
"Start and end of expansion must be
in the same ultimate file!"); <br>
<br>
bool stopOutputOnNextToken = false; <br>
bool toReplaceStartsInMacro =
toReplaceExpStart == TokExp; <br>
bool toReplaceEndsInMacro =
toReplaceExpEnd == TokExp; <br>
bool startedOutput = false; <br>
<br>
Token PrevPrevTok; <br>
Token PrevTok = Tok; <br>
<br>
while (!Tok.is(tok::eof) &&
SM.getExpansionLoc(Tok.getLocation()) == LLoc.first) <br>
{ <br>
printf("TOKEN (in macro): %s\n",
TmpPP.getSpelling(Tok).c_str()); <br>
<br>
auto TokSpell =
SM.getSpellingLoc(Tok.getLocation()); <br>
if (stopOutputOnNextToken) <br>
{ <br>
break; <br>
} <br>
if (toReplaceEndsInMacro &&
TokSpell == toReplaceSpellEnd) <br>
{ <br>
stopOutputOnNextToken = true; <br>
} <br>
<br>
if (toReplaceStartsInMacro &&
!startedOutput) <br>
{ <br>
if (TokSpell ==
toReplaceSpellStart) <br>
{ <br>
startedOutput = true; <br>
} <br>
else <br>
{ <br>
TmpPP.Lex(Tok); <br>
continue; <br>
} <br>
} <br>
<br>
// If the tokens were already space
separated, or if they must be to avoid <br>
// them being implicitly pasted, add a
space between them. <br>
if (Tok.hasLeadingSpace() ||
ConcatInfo.AvoidConcat(PrevPrevTok, PrevTok, Tok)) <br>
out += ' '; <br>
<br>
if (checkReplacement()) <br>
{ <br>
continue; <br>
} <br>
<br>
out += TmpPP.getSpelling(Tok); <br>
TmpPP.Lex(Tok); <br>
} <br>
if (stopOutputOnNextToken) <br>
{ <br>
break; <br>
} <br>
} <br>
else <br>
{ <br>
if (checkReplacement()) <br>
{ <br>
continue; <br>
} <br>
<br>
// Output original code because we are
outside of a replacement. <br>
out += TmpPP.getSpelling(Tok); <br>
TmpPP.Lex(Tok); <br>
} <br>
} <br>
else <br>
{ <br>
TmpPP.Lex(Tok); <br>
} <br>
} <br>
<br>
// Restore the preprocessor's old state. <br>
TmpPP.setDiagnostics(*OldDiags); <br>
TmpPP.setPragmasEnabled(PragmasPreviouslyEnabled); <br>
<br>
for (const auto &tmpm : TmpPP.macros()) <br>
{ <br>
auto it = MacroPreviouslyEnabled.find(tmpm.first);
<br>
if (it != MacroPreviouslyEnabled.end()) <br>
{ <br>
auto MD = tmpm.second.getLatest(); <br>
auto MI = MD->getMacroInfo(); <br>
if (MI->isEnabled() &&
!it->second) <br>
{ <br>
MI->DisableMacro(); <br>
} <br>
else if (!MI->isEnabled() &&
it->second) <br>
{ <br>
MI->EnableMacro(); <br>
} <br>
} <br>
} <br>
<br>
return out; <br>
} <br>
<br>
<br>
<fieldset class="m_-273721086631250889mimeAttachmentHeader"></fieldset>
<br>
<pre>_______________________________________________
cfe-dev mailing list
<a class="m_-273721086631250889moz-txt-link-abbreviated" href="mailto:cfe-dev@lists.llvm.org" target="_blank">cfe-dev@lists.llvm.org</a>
<a class="m_-273721086631250889moz-txt-link-freetext" href="http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev</a>
</pre>
</blockquote>
<br>
<br>
<fieldset class="m_-273721086631250889mimeAttachmentHeader"></fieldset>
<br>
<pre>_______________________________________________
cfe-dev mailing list
<a class="m_-273721086631250889moz-txt-link-abbreviated" href="mailto:cfe-dev@lists.llvm.org" target="_blank">cfe-dev@lists.llvm.org</a>
<a class="m_-273721086631250889moz-txt-link-freetext" href="http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev</a>
</pre>
</blockquote>
<br>
</blockquote>
<br>
</blockquote>
</blockquote>
</div>
_______________________________________________<br>
cfe-dev mailing list<br>
<a href="mailto:cfe-dev@lists.llvm.org" target="_blank">cfe-dev@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev</a><br>
</blockquote></div>