skia2/tools/bookmaker/includeWriter.cpp
Cary Clark 154beea859 Add docs for SkMatrix, SkRect, SkIRect, SkBitmap
Also minor changes to earlier docs.

Many small changes to improve indentation in generated includes.
Added support for matrix math illustrations.

Docs-Preview: https://skia.org/?cl=58500
Bug: skia:6898
Change-Id: I7da58ad55f82d7fd41d19288beb2cd71730fb01f
Reviewed-on: https://skia-review.googlesource.com/58500
Commit-Queue: Cary Clark <caryclark@skia.org>
Reviewed-by: Cary Clark <caryclark@google.com>
Reviewed-by: Cary Clark <caryclark@skia.org>
2017-10-26 12:17:36 +00:00

1928 lines
77 KiB
C++

/*
* Copyright 2017 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "bookmaker.h"
void IncludeWriter::descriptionOut(const Definition* def) {
const char* commentStart = def->fContentStart;
int commentLen = (int) (def->fContentEnd - commentStart);
bool breakOut = false;
SkDEBUGCODE(bool wroteCode = false);
for (auto prop : def->fChildren) {
switch (prop->fMarkType) {
case MarkType::kCode: {
bool literal = false;
bool literalOutdent = false;
commentLen = (int) (prop->fStart - commentStart);
if (commentLen > 0) {
SkASSERT(commentLen < 1000);
if (Wrote::kNone != this->rewriteBlock(commentLen, commentStart, Phrase::kNo)) {
this->lf(2);
}
}
size_t childSize = prop->fChildren.size();
if (childSize) {
SkASSERT(1 == childSize || 2 == childSize); // incomplete
SkASSERT(MarkType::kLiteral == prop->fChildren[0]->fMarkType);
SkASSERT(1 == childSize || MarkType::kOutdent == prop->fChildren[1]->fMarkType);
commentStart = prop->fChildren[childSize - 1]->fContentStart;
literal = true;
literalOutdent = 2 == childSize &&
MarkType::kOutdent == prop->fChildren[1]->fMarkType;
}
commentLen = (int) (prop->fContentEnd - commentStart);
SkASSERT(commentLen > 0);
if (literal) {
if (!literalOutdent) {
fIndent += 4;
}
this->writeBlockIndent(commentLen, commentStart);
this->lf(2);
if (!literalOutdent) {
fIndent -= 4;
}
commentStart = prop->fTerminator;
SkDEBUGCODE(wroteCode = true);
}
} break;
case MarkType::kDefinedBy:
commentStart = prop->fTerminator;
break;
case MarkType::kDeprecated:
case MarkType::kPrivate:
commentLen = (int) (prop->fStart - commentStart);
if (commentLen > 0) {
SkASSERT(commentLen < 1000);
if (Wrote::kNone != this->rewriteBlock(commentLen, commentStart, Phrase::kNo)) {
this->lfcr();
}
}
commentStart = prop->fContentStart;
commentLen = (int) (prop->fContentEnd - commentStart);
if (commentLen > 0) {
if (Wrote::kNone != this->rewriteBlock(commentLen, commentStart, Phrase::kNo)) {
this->lfcr();
}
}
commentStart = prop->fTerminator;
commentLen = (int) (def->fContentEnd - commentStart);
break;
case MarkType::kExperimental:
this->writeString("EXPERIMENTAL:");
this->writeSpace();
commentStart = prop->fContentStart;
commentLen = (int) (prop->fContentEnd - commentStart);
if (commentLen > 0) {
if (Wrote::kNone != this->rewriteBlock(commentLen, commentStart, Phrase::kNo)) {
this->lfcr();
}
}
commentStart = prop->fTerminator;
commentLen = (int) (def->fContentEnd - commentStart);
break;
case MarkType::kFormula: {
commentLen = prop->fStart - commentStart;
if (commentLen > 0) {
if (Wrote::kNone != this->rewriteBlock(commentLen, commentStart, Phrase::kNo)) {
if (commentLen > 1 && '\n' == prop->fStart[-1] &&
'\n' == prop->fStart[-2]) {
this->lf(1);
} else {
this->writeSpace();
}
}
}
int saveIndent = fIndent;
if (fIndent < fColumn + 1) {
fIndent = fColumn + 1;
}
this->writeBlockIndent(prop->length(), prop->fContentStart);
fIndent = saveIndent;
commentStart = prop->fTerminator;
commentLen = (int) (def->fContentEnd - commentStart);
if (commentLen > 1 && '\n' == commentStart[0] && '\n' == commentStart[1]) {
this->lf(2);
} else {
SkASSERT('\n' == prop->fTerminator[0]);
if ('.' != prop->fTerminator[1] && !fLinefeeds) {
this->writeSpace();
}
}
} break;
case MarkType::kToDo:
commentLen = (int) (prop->fStart - commentStart);
if (commentLen > 0) {
SkASSERT(commentLen < 1000);
if (Wrote::kNone != this->rewriteBlock(commentLen, commentStart, Phrase::kNo)) {
this->lfcr();
}
}
commentStart = prop->fTerminator;
commentLen = (int) (def->fContentEnd - commentStart);
break;
case MarkType::kList:
commentLen = prop->fStart - commentStart;
if (commentLen > 0) {
if (Wrote::kNone != this->rewriteBlock(commentLen, commentStart,
Phrase::kNo)) {
this->lfcr();
}
}
for (auto row : prop->fChildren) {
SkASSERT(MarkType::kRow == row->fMarkType);
for (auto column : row->fChildren) {
SkASSERT(MarkType::kColumn == column->fMarkType);
this->writeString("-");
this->writeSpace();
this->descriptionOut(column);
this->lf(1);
}
}
commentStart = prop->fTerminator;
commentLen = (int) (def->fContentEnd - commentStart);
if ('\n' == commentStart[0] && '\n' == commentStart[1]) {
this->lf(2);
}
break;
default:
commentLen = (int) (prop->fStart - commentStart);
breakOut = true;
}
if (breakOut) {
break;
}
}
SkASSERT(wroteCode || (commentLen > 0 && commentLen < 1500));
if (commentLen > 0) {
this->rewriteBlock(commentLen, commentStart, Phrase::kNo);
}
}
void IncludeWriter::enumHeaderOut(const RootDefinition* root,
const Definition& child) {
const Definition* enumDef = nullptr;
const char* bodyEnd = fDeferComment ? fDeferComment->fContentStart - 1 :
child.fContentStart;
this->writeBlockTrim((int) (bodyEnd - fStart), fStart); // may write nothing
this->lf(2);
if (fIndentNext) {
fIndent += 4;
fIndentNext = false;
}
fDeferComment = nullptr;
fStart = child.fContentStart;
const auto& nameDef = child.fTokens.front();
string fullName;
if (nullptr != nameDef.fContentEnd) {
TextParser enumClassCheck(&nameDef);
const char* start = enumClassCheck.fStart;
size_t len = (size_t) (enumClassCheck.fEnd - start);
bool enumClass = enumClassCheck.skipExact("class ");
if (enumClass) {
start = enumClassCheck.fChar;
const char* end = enumClassCheck.anyOf(" \n;{");
len = (size_t) (end - start);
}
string enumName(start, len);
if (enumClass) {
child.fChildren[0]->fName = enumName;
}
fullName = root->fName + "::" + enumName;
enumDef = root->find(enumName, RootDefinition::AllowParens::kNo);
if (!enumDef) {
enumDef = root->find(fullName, RootDefinition::AllowParens::kNo);
}
SkASSERT(enumDef);
// child[0] should be #Code comment starts at child[0].fTerminator
// though skip until #Code is found (in case there's a #ToDo, etc)
// child[1] should be #Const comment ends at child[1].fStart
// comment becomes enum header (if any)
} else {
string enumName(root->fName);
enumName += "::_anonymous";
if (fAnonymousEnumCount > 1) {
enumName += '_' + to_string(fAnonymousEnumCount);
}
enumDef = root->find(enumName, RootDefinition::AllowParens::kNo);
SkASSERT(enumDef);
++fAnonymousEnumCount;
}
Definition* codeBlock = nullptr;
const char* commentStart = nullptr;
bool wroteHeader = false;
bool lastAnchor = false;
SkDEBUGCODE(bool foundConst = false);
for (auto test : enumDef->fChildren) {
if (MarkType::kCode == test->fMarkType) {
SkASSERT(!codeBlock); // FIXME: check enum for correct order earlier
codeBlock = test;
commentStart = codeBlock->fTerminator;
continue;
}
if (!codeBlock) {
continue;
}
const char* commentEnd = test->fStart;
if (!wroteHeader &&
!this->contentFree((int) (commentEnd - commentStart), commentStart)) {
if (fIndentNext) {
fIndent += 4;
}
this->writeCommentHeader();
this->writeString("\\enum");
if (fullName.length() > 0) {
this->writeSpace();
this->writeString(fullName.c_str());
}
fIndent += 4;
this->lfcr();
wroteHeader = true;
}
if (lastAnchor) {
if (commentEnd - commentStart > 1) {
SkASSERT('\n' == commentStart[0]);
if (' ' == commentStart[1]) {
this->writeSpace();
}
}
lastAnchor = false;
}
this->rewriteBlock((int) (commentEnd - commentStart), commentStart, Phrase::kNo);
if (MarkType::kAnchor == test->fMarkType) {
bool newLine = commentEnd - commentStart > 1 &&
'\n' == commentEnd[-1] && '\n' == commentEnd[-2];
commentStart = test->fContentStart;
commentEnd = test->fChildren[0]->fStart;
if (newLine) {
this->lf(2);
} else {
this->writeSpace();
}
this->rewriteBlock((int) (commentEnd - commentStart), commentStart, Phrase::kNo);
lastAnchor = true; // this->writeSpace();
}
commentStart = test->fTerminator;
if (MarkType::kConst == test->fMarkType) {
SkASSERT(codeBlock); // FIXME: check enum for correct order earlier
SkDEBUGCODE(foundConst = true);
break;
}
}
SkASSERT(codeBlock);
SkASSERT(foundConst);
if (wroteHeader) {
fIndent -= 4;
this->lfcr();
this->writeCommentTrailer();
}
Definition* braceHolder = child.fChildren[0];
if (KeyWord::kClass == braceHolder->fKeyWord) {
braceHolder = braceHolder->fChildren[0];
}
bodyEnd = braceHolder->fContentStart;
SkASSERT('{' == bodyEnd[0]);
++bodyEnd;
this->lfcr();
this->writeBlock((int) (bodyEnd - fStart), fStart); // write include "enum Name {"
fIndent += 4;
this->singleLF();
fStart = bodyEnd;
fEnumDef = enumDef;
}
void IncludeWriter::enumMembersOut(const RootDefinition* root, Definition& child) {
// iterate through include tokens and find how much remains for 1 line comments
// put ones that fit on same line, ones that are too big on preceding line?
const Definition* currentEnumItem = nullptr;
const char* commentStart = nullptr;
const char* lastEnd = nullptr;
int commentLen = 0;
enum class State {
kNoItem,
kItemName,
kItemValue,
kItemComment,
};
State state = State::kNoItem;
vector<IterState> iterStack;
iterStack.emplace_back(child.fTokens.begin(), child.fTokens.end());
IterState* iterState = &iterStack[0];
bool preprocessorWord = false;
const char* preprocessStart = nullptr;
const char* preprocessEnd = nullptr;
for (int onePast = 0; onePast < 2; onePast += iterState->fDefIter == iterState->fDefEnd) {
Definition* token = onePast ? nullptr : &*iterState->fDefIter++;
if (token && Definition::Type::kBracket == token->fType) {
if (Bracket::kSlashSlash == token->fBracket) {
fStart = token->fContentEnd;
continue; // ignore old inline comments
}
if (Bracket::kSlashStar == token->fBracket) {
fStart = token->fContentEnd + 1;
continue; // ignore old inline comments
}
if (Bracket::kPound == token->fBracket) { // preprocessor wraps member
preprocessStart = token->fContentStart;
if (KeyWord::kIf == token->fKeyWord || KeyWord::kIfdef == token->fKeyWord) {
iterStack.emplace_back(token->fTokens.begin(), token->fTokens.end());
iterState = &iterStack.back();
preprocessorWord = true;
} else if (KeyWord::kEndif == token->fKeyWord) {
iterStack.pop_back();
iterState = &iterStack.back();
preprocessEnd = token->fContentEnd;
} else {
SkASSERT(0); // incomplete
}
continue;
}
SkASSERT(0); // incomplete
}
if (token && Definition::Type::kWord != token->fType) {
SkASSERT(0); // incomplete
}
if (preprocessorWord) {
preprocessorWord = false;
preprocessEnd = token->fContentEnd;
continue;
}
if (token && State::kItemName == state) {
TextParser enumLine(token->fFileName, lastEnd,
token->fContentStart, token->fLineCount);
const char* end = enumLine.anyOf(",}=");
SkASSERT(end);
state = '=' == *end ? State::kItemValue : State::kItemComment;
if (State::kItemValue == state) { // write enum value
this->indentToColumn(fEnumItemValueTab);
this->writeString("=");
this->writeSpace();
lastEnd = token->fContentEnd;
this->writeBlock((int) (lastEnd - token->fContentStart),
token->fContentStart); // write const value if any
continue;
}
}
if (token && State::kItemValue == state) {
TextParser valueEnd(token->fFileName, lastEnd,
token->fContentStart, token->fLineCount);
const char* end = valueEnd.anyOf(",}");
if (!end) { // write expression continuation
if (' ' == lastEnd[0]) {
this->writeSpace();
}
this->writeBlock((int) (token->fContentEnd - lastEnd), lastEnd);
continue;
}
}
if (State::kNoItem != state) {
this->writeString(",");
SkASSERT(currentEnumItem);
if (currentEnumItem->fShort) {
this->indentToColumn(fEnumItemCommentTab);
this->writeString("//!<");
this->writeSpace();
this->rewriteBlock(commentLen, commentStart, Phrase::kNo);
}
if (onePast) {
fIndent -= 4;
}
this->lfcr();
if (preprocessStart) {
SkASSERT(preprocessEnd);
int saveIndent = fIndent;
fIndent = SkTMax(0, fIndent - 8);
this->lf(2);
this->writeBlock((int) (preprocessEnd - preprocessStart), preprocessStart);
this->lfcr();
fIndent = saveIndent;
preprocessStart = nullptr;
preprocessEnd = nullptr;
}
if (token && State::kItemValue == state) {
fStart = token->fContentStart;
}
state = State::kNoItem;
}
SkASSERT(State::kNoItem == state);
if (onePast) {
break;
}
SkASSERT(token);
string itemName = root->fName + "::";
if (KeyWord::kClass == child.fParent->fKeyWord) {
itemName += child.fParent->fName + "::";
}
itemName += string(token->fContentStart, (int) (token->fContentEnd - token->fContentStart));
for (auto& enumItem : fEnumDef->fChildren) {
if (MarkType::kConst != enumItem->fMarkType) {
continue;
}
if (itemName != enumItem->fName) {
continue;
}
currentEnumItem = enumItem;
break;
}
SkASSERT(currentEnumItem);
// if description fits, it goes after item
commentStart = currentEnumItem->fContentStart;
const char* commentEnd;
if (currentEnumItem->fChildren.size() > 0) {
commentEnd = currentEnumItem->fChildren[0]->fStart;
} else {
commentEnd = currentEnumItem->fContentEnd;
}
TextParser enumComment(fFileName, commentStart, commentEnd, currentEnumItem->fLineCount);
if (enumComment.skipToLineStart()) { // skip const value
commentStart = enumComment.fChar;
commentLen = (int) (commentEnd - commentStart);
} else {
const Definition* privateDef = currentEnumItem->fChildren[0];
SkASSERT(MarkType::kPrivate == privateDef->fMarkType);
commentStart = privateDef->fContentStart;
commentLen = (int) (privateDef->fContentEnd - privateDef->fContentStart);
}
// FIXME: may assert here if there's no const value
// should have detected and errored on that earlier when enum fContentStart was set
SkASSERT(commentLen > 0 && commentLen < 1000);
if (!currentEnumItem->fShort) {
this->writeCommentHeader();
fIndent += 4;
bool wroteLineFeed = Wrote::kLF ==
this->rewriteBlock(commentLen, commentStart, Phrase::kNo);
fIndent -= 4;
if (wroteLineFeed || fColumn > 100 - 3 /* space * / */ ) {
this->lfcr();
} else {
this->writeSpace();
}
this->writeCommentTrailer();
}
lastEnd = token->fContentEnd;
this->lfcr();
if (',' == fStart[0]) {
++fStart;
}
this->writeBlock((int) (lastEnd - fStart), fStart); // enum item name
fStart = token->fContentEnd;
state = State::kItemName;
}
}
void IncludeWriter::enumSizeItems(const Definition& child) {
enum class State {
kNoItem,
kItemName,
kItemValue,
kItemComment,
};
State state = State::kNoItem;
int longestName = 0;
int longestValue = 0;
int valueLen = 0;
const char* lastEnd = nullptr;
SkASSERT(child.fChildren.size() == 1 || child.fChildren.size() == 2);
auto brace = child.fChildren[0];
if (KeyWord::kClass == brace->fKeyWord) {
brace = brace->fChildren[0];
}
SkASSERT(Bracket::kBrace == brace->fBracket);
vector<IterState> iterStack;
iterStack.emplace_back(brace->fTokens.begin(), brace->fTokens.end());
IterState* iterState = &iterStack[0];
bool preprocessorWord = false;
while (iterState->fDefIter != iterState->fDefEnd) {
auto& token = *iterState->fDefIter++;
if (Definition::Type::kBracket == token.fType) {
if (Bracket::kSlashSlash == token.fBracket) {
continue; // ignore old inline comments
}
if (Bracket::kSlashStar == token.fBracket) {
continue; // ignore old inline comments
}
if (Bracket::kPound == token.fBracket) { // preprocessor wraps member
if (KeyWord::kIf == token.fKeyWord || KeyWord::kIfdef == token.fKeyWord) {
iterStack.emplace_back(token.fTokens.begin(), token.fTokens.end());
iterState = &iterStack.back();
preprocessorWord = true;
} else if (KeyWord::kEndif == token.fKeyWord) {
iterStack.pop_back();
iterState = &iterStack.back();
} else {
SkASSERT(0); // incomplete
}
continue;
}
SkASSERT(0); // incomplete
}
if (Definition::Type::kWord != token.fType) {
SkASSERT(0); // incomplete
}
if (preprocessorWord) {
preprocessorWord = false;
continue;
}
if (State::kItemName == state) {
TextParser enumLine(token.fFileName, lastEnd,
token.fContentStart, token.fLineCount);
const char* end = enumLine.anyOf(",}=");
SkASSERT(end);
state = '=' == *end ? State::kItemValue : State::kItemComment;
if (State::kItemValue == state) {
valueLen = (int) (token.fContentEnd - token.fContentStart);
lastEnd = token.fContentEnd;
continue;
}
}
if (State::kItemValue == state) {
TextParser valueEnd(token.fFileName, lastEnd,
token.fContentStart, token.fLineCount);
const char* end = valueEnd.anyOf(",}");
if (!end) { // write expression continuation
valueLen += (int) (token.fContentEnd - lastEnd);
continue;
}
}
if (State::kNoItem != state) {
longestValue = SkTMax(longestValue, valueLen);
state = State::kNoItem;
}
SkASSERT(State::kNoItem == state);
lastEnd = token.fContentEnd;
longestName = SkTMax(longestName, (int) (lastEnd - token.fContentStart));
state = State::kItemName;
}
if (State::kItemValue == state) {
longestValue = SkTMax(longestValue, valueLen);
}
fEnumItemValueTab = longestName + fIndent + 1 /* space before = */ ;
if (longestValue) {
longestValue += 3; /* = space , */
}
fEnumItemCommentTab = fEnumItemValueTab + longestValue + 1 /* space before //!< */ ;
// iterate through bmh children and see which comments fit on include lines
for (auto& enumItem : fEnumDef->fChildren) {
if (MarkType::kConst != enumItem->fMarkType) {
continue;
}
TextParser enumLine(enumItem);
enumLine.trimEnd();
enumLine.skipToLineStart(); // skip const value
const char* commentStart = enumLine.fChar;
enumLine.skipLine();
ptrdiff_t lineLen = enumLine.fChar - commentStart + 5 /* //!< space */ ;
if (!enumLine.eof()) {
enumLine.skipWhiteSpace();
}
enumItem->fShort = enumLine.eof() && fEnumItemCommentTab + lineLen < 100;
}
}
// walk children and output complete method doxygen description
void IncludeWriter::methodOut(const Definition* method, const Definition& child) {
if (fPendingMethod) {
fIndent -= 4;
fPendingMethod = false;
}
fBmhMethod = method;
fMethodDef = &child;
fContinuation = nullptr;
fDeferComment = nullptr;
if (0 == fIndent || fIndentNext) {
fIndent += 4;
fIndentNext = false;
}
this->writeCommentHeader();
fIndent += 4;
this->descriptionOut(method);
// compute indention column
size_t column = 0;
bool hasParmReturn = false;
for (auto methodPart : method->fChildren) {
if (MarkType::kParam == methodPart->fMarkType) {
column = SkTMax(column, methodPart->fName.length());
hasParmReturn = true;
} else if (MarkType::kReturn == methodPart->fMarkType) {
hasParmReturn = true;
}
}
if (hasParmReturn) {
this->lf(2);
column += fIndent + sizeof("@return ");
int saveIndent = fIndent;
for (auto methodPart : method->fChildren) {
const char* partStart = methodPart->fContentStart;
const char* partEnd = methodPart->fContentEnd;
if (MarkType::kParam == methodPart->fMarkType) {
this->writeString("@param");
this->writeSpace();
this->writeString(methodPart->fName.c_str());
} else if (MarkType::kReturn == methodPart->fMarkType) {
this->writeString("@return");
} else {
continue;
}
while ('\n' == partEnd[-1]) {
--partEnd;
}
while ('#' == partEnd[-1]) { // FIXME: so wrong; should not be before fContentEnd
--partEnd;
}
this->indentToColumn(column);
int partLen = (int) (partEnd - partStart);
// FIXME : detect this earlier; assert if #Return is empty
SkASSERT(partLen > 0 && partLen < 200);
fIndent = column;
this->rewriteBlock(partLen, partStart, Phrase::kYes);
fIndent = saveIndent;
this->lfcr();
}
} else {
this->lfcr();
}
fIndent -= 4;
this->lfcr();
this->writeCommentTrailer();
fBmhMethod = nullptr;
fMethodDef = nullptr;
fWroteMethod = true;
}
void IncludeWriter::structOut(const Definition* root, const Definition& child,
const char* commentStart, const char* commentEnd) {
this->writeCommentHeader();
this->writeString("\\");
SkASSERT(MarkType::kClass == child.fMarkType || MarkType::kStruct == child.fMarkType);
this->writeString(MarkType::kClass == child.fMarkType ? "class" : "struct");
this->writeSpace();
this->writeString(child.fName.c_str());
fIndent += 4;
this->lfcr();
this->rewriteBlock((int) (commentEnd - commentStart), commentStart, Phrase::kNo);
fIndent -= 4;
this->lfcr();
this->writeCommentTrailer();
}
Definition* IncludeWriter::structMemberOut(const Definition* memberStart, const Definition& child) {
const char* blockStart = !fWroteMethod && fDeferComment ? fLastComment->fContentEnd : fStart;
const char* blockEnd = fWroteMethod && fDeferComment ? fDeferComment->fStart - 1 :
memberStart->fStart;
this->writeBlockTrim((int) (blockEnd - blockStart), blockStart);
if (fIndentNext) {
fIndent += 4;
fIndentNext = false;
}
fWroteMethod = false;
const char* commentStart = nullptr;
ptrdiff_t commentLen = 0;
string name(child.fContentStart, (int) (child.fContentEnd - child.fContentStart));
bool isShort;
Definition* commentBlock = nullptr;
for (auto memberDef : fBmhStructDef->fChildren) {
if (memberDef->fName.length() - name.length() == memberDef->fName.find(name)) {
commentStart = memberDef->fContentStart;
commentLen = memberDef->fContentEnd - commentStart;
isShort = memberDef->fShort;
commentBlock = memberDef;
SkASSERT(!isShort || memberDef->fChildren.size() == 0);
break;
}
}
if (!isShort) {
this->writeCommentHeader();
bool wroteLineFeed = false;
fIndent += 4;
for (auto child : commentBlock->fChildren) {
commentLen = child->fStart - commentStart;
wroteLineFeed |= Wrote::kLF == this->rewriteBlock(commentLen, commentStart, Phrase::kNo);
if (MarkType::kFormula == child->fMarkType) {
this->writeSpace();
this->writeBlock((int) (child->fContentEnd - child->fContentStart),
child->fContentStart);
}
commentStart = child->fTerminator;
}
commentLen = commentBlock->fContentEnd - commentStart;
wroteLineFeed |= Wrote::kLF == this->rewriteBlock(commentLen, commentStart, Phrase::kNo);
fIndent -= 4;
if (wroteLineFeed || fColumn > 100 - 3 /* space * / */ ) {
this->lfcr();
} else {
this->writeSpace();
}
this->writeCommentTrailer();
}
this->lfcr();
this->writeBlock((int) (child.fStart - memberStart->fContentStart),
memberStart->fContentStart);
this->indentToColumn(fStructMemberTab);
this->writeString(name.c_str());
auto tokenIter = child.fParent->fTokens.begin();
std::advance(tokenIter, child.fParentIndex + 1);
Definition* valueStart = &*tokenIter;
while (Definition::Type::kPunctuation != tokenIter->fType) {
std::advance(tokenIter, 1);
SkASSERT(child.fParent->fTokens.end() != tokenIter);
}
Definition* valueEnd = &*tokenIter;
if (valueStart != valueEnd) {
this->indentToColumn(fStructValueTab);
this->writeString("=");
this->writeSpace();
this->writeBlock((int) (valueEnd->fStart - valueStart->fContentStart),
valueStart->fContentStart);
}
this->writeString(";");
if (isShort) {
this->indentToColumn(fStructCommentTab);
this->writeString("//!<");
this->writeSpace();
this->rewriteBlock(commentLen, commentStart, Phrase::kNo);
}
this->lf(2);
return valueEnd;
}
void IncludeWriter::structSizeMembers(const Definition& child) {
int longestType = 0;
Definition* typeStart = nullptr;
int longestName = 0;
int longestValue = 0;
SkASSERT(child.fChildren.size() == 1 || child.fChildren.size() == 2);
bool inEnum = false;
bool inMethod = false;
bool inMember = false;
auto brace = child.fChildren[0];
SkASSERT(Bracket::kBrace == brace->fBracket);
for (auto& token : brace->fTokens) {
if (Definition::Type::kBracket == token.fType) {
if (Bracket::kSlashSlash == token.fBracket) {
continue; // ignore old inline comments
}
if (Bracket::kSlashStar == token.fBracket) {
continue; // ignore old inline comments
}
if (Bracket::kParen == token.fBracket) {
if (inMethod) {
continue;
}
break;
}
SkASSERT(0); // incomplete
}
if (Definition::Type::kKeyWord == token.fType) {
switch (token.fKeyWord) {
case KeyWord::kEnum:
inEnum = true;
break;
case KeyWord::kConst:
case KeyWord::kConstExpr:
case KeyWord::kStatic:
case KeyWord::kInt:
case KeyWord::kUint8_t:
case KeyWord::kUint16_t:
case KeyWord::kUint32_t:
case KeyWord::kUint64_t:
case KeyWord::kSize_t:
case KeyWord::kFloat:
case KeyWord::kBool:
case KeyWord::kVoid:
if (!typeStart) {
typeStart = &token;
}
break;
default:
break;
}
continue;
}
if (Definition::Type::kPunctuation == token.fType) {
if (inEnum) {
SkASSERT(Punctuation::kSemicolon == token.fPunctuation);
inEnum = false;
}
if (inMethod) {
if (Punctuation::kColon == token.fPunctuation) {
inMethod = false;
} else if (Punctuation::kLeftBrace == token.fPunctuation) {
inMethod = false;
} else if (Punctuation::kSemicolon == token.fPunctuation) {
inMethod = false;
} else {
SkASSERT(0); // incomplete
}
}
if (inMember) {
SkASSERT(Punctuation::kSemicolon == token.fPunctuation);
typeStart = nullptr;
inMember = false;
}
continue;
}
if (Definition::Type::kWord != token.fType) {
SkASSERT(0); // incomplete
}
if (MarkType::kMember == token.fMarkType) {
TextParser typeStr(token.fFileName, typeStart->fContentStart, token.fContentStart,
token.fLineCount);
typeStr.trimEnd();
longestType = SkTMax(longestType, (int) (typeStr.fEnd - typeStr.fStart));
longestName = SkTMax(longestName, (int) (token.fContentEnd - token.fContentStart));
typeStart->fMemberStart = true;
inMember = true;
continue;
}
if (MarkType::kMethod == token.fMarkType) {
inMethod = true;
continue;
}
SkASSERT(MarkType::kNone == token.fMarkType);
if (typeStart) {
if (inMember) {
longestValue =
SkTMax(longestValue, (int) (token.fContentEnd - token.fContentStart));
}
} else {
typeStart = &token;
}
}
fStructMemberTab = longestType + fIndent + 1 /* space before name */ ;
fStructValueTab = fStructMemberTab + longestName + 2 /* space ; */ ;
fStructCommentTab = fStructValueTab;
if (longestValue) {
fStructCommentTab += longestValue + 3 /* space = space */ ;
fStructValueTab -= 1 /* ; */ ;
}
// iterate through bmh children and see which comments fit on include lines
for (auto& member : fBmhStructDef->fChildren) {
if (MarkType::kMember != member->fMarkType) {
continue;
}
TextParser memberLine(member);
memberLine.trimEnd();
const char* commentStart = memberLine.fChar;
memberLine.skipLine();
ptrdiff_t lineLen = memberLine.fChar - commentStart + 5 /* //!< space */ ;
if (!memberLine.eof()) {
memberLine.skipWhiteSpace();
}
member->fShort = memberLine.eof() && fStructCommentTab + lineLen < 100;
}
}
static bool find_start(const Definition* startDef, const char* start) {
for (const auto& child : startDef->fTokens) {
if (child.fContentStart == start) {
return MarkType::kMethod == child.fMarkType;
}
if (child.fContentStart >= start) {
break;
}
if (find_start(&child, start)) {
return true;
}
}
return false;
}
bool IncludeWriter::populate(Definition* def, ParentPair* prevPair, RootDefinition* root) {
ParentPair pair = { def, prevPair };
// write bulk of original include up to class, method, enum, etc., excepting preceding comment
// find associated bmh object
// write any associated comments in Doxygen form
// skip include comment
// if there is a series of same named methods, write one set of comments, then write all methods
string methodName;
const Definition* method;
const Definition* clonedMethod = nullptr;
const Definition* memberStart = nullptr;
const Definition* memberEnd = nullptr;
fContinuation = nullptr;
bool inStruct = false;
bool inConstructor = false;
bool inInline = false;
bool eatOperator = false;
const Definition* requireDense = nullptr;
const Definition* startDef = nullptr;
for (auto& child : def->fTokens) {
if (KeyWord::kOperator == child.fKeyWord && method &&
Definition::MethodType::kOperator == method->fMethodType) {
eatOperator = true;
continue;
}
if (eatOperator) {
if (Bracket::kSquare == child.fBracket || Bracket::kParen == child.fBracket) {
continue;
}
eatOperator = false;
fContinuation = nullptr;
if (KeyWord::kConst == child.fKeyWord) {
continue;
}
}
if (memberEnd) {
if (memberEnd != &child) {
continue;
}
startDef = &child;
fStart = child.fContentStart + 1;
memberEnd = nullptr;
}
if (child.fPrivate) {
if (MarkType::kMethod == child.fMarkType) {
inInline = true;
}
continue;
}
if (inInline) {
if (Definition::Type::kKeyWord == child.fType) {
SkASSERT(MarkType::kMethod != child.fMarkType);
continue;
}
if (Definition::Type::kPunctuation == child.fType) {
if (Punctuation::kLeftBrace == child.fPunctuation) {
inInline = false;
} else {
SkASSERT(Punctuation::kAsterisk == child.fPunctuation);
}
continue;
}
if (Definition::Type::kWord == child.fType) {
string name(child.fContentStart, child.fContentEnd - child.fContentStart);
SkASSERT(string::npos != name.find("::"));
continue;
}
if (Definition::Type::kBracket == child.fType) {
SkASSERT(Bracket::kParen == child.fBracket);
continue;
}
}
if (fContinuation) {
if (Definition::Type::kKeyWord == child.fType) {
if (KeyWord::kFriend == child.fKeyWord ||
KeyWord::kSK_API == child.fKeyWord) {
continue;
}
const IncludeKey& includeKey = kKeyWords[(int) child.fKeyWord];
if (KeyProperty::kNumber == includeKey.fProperty) {
continue;
}
}
if (Definition::Type::kBracket == child.fType && Bracket::kParen == child.fBracket) {
if (!clonedMethod) {
if (inConstructor) {
fContinuation = child.fContentStart;
}
continue;
}
int alternate = 1;
ptrdiff_t childLen = child.fContentEnd - child.fContentStart;
SkASSERT(')' == child.fContentStart[childLen]);
++childLen;
do {
TextParser params(clonedMethod->fFileName, clonedMethod->fStart,
clonedMethod->fContentStart, clonedMethod->fLineCount);
params.skipToEndBracket('(');
if (params.startsWith(child.fContentStart, childLen)) {
this->methodOut(clonedMethod, child);
break;
}
++alternate;
string alternateMethod = methodName + '_' + to_string(alternate);
clonedMethod = root->find(alternateMethod, RootDefinition::AllowParens::kNo);
} while (clonedMethod);
if (!clonedMethod) {
return this->reportError<bool>("cloned method not found");
}
clonedMethod = nullptr;
continue;
}
if (Definition::Type::kWord == child.fType) {
if (clonedMethod) {
continue;
}
size_t len = (size_t) (child.fContentEnd - child.fContentStart);
const char operatorStr[] = "operator";
size_t operatorLen = sizeof(operatorStr) - 1;
if (len >= operatorLen && !strncmp(child.fContentStart, operatorStr, operatorLen)) {
fContinuation = child.fContentEnd;
continue;
}
}
if (Definition::Type::kPunctuation == child.fType &&
(Punctuation::kSemicolon == child.fPunctuation ||
Punctuation::kLeftBrace == child.fPunctuation ||
(Punctuation::kColon == child.fPunctuation && inConstructor))) {
SkASSERT(fContinuation[0] == '(');
const char* continueEnd = child.fContentStart;
while (continueEnd > fContinuation && isspace(continueEnd[-1])) {
--continueEnd;
}
methodName += string(fContinuation, continueEnd - fContinuation);
method = root->find(methodName, RootDefinition::AllowParens::kNo);
if (!method) {
fLineCount = child.fLineCount;
return this->reportError<bool>("method not found");
}
this->methodOut(method, child);
continue;
}
if (Definition::Type::kPunctuation == child.fType &&
Punctuation::kAsterisk == child.fPunctuation &&
clonedMethod) {
continue;
}
if (inConstructor) {
continue;
}
method = root->find(methodName + "()", RootDefinition::AllowParens::kNo);
if (method && MarkType::kDefinedBy == method->fMarkType) {
method = method->fParent;
}
if (method) {
if (method->fCloned) {
clonedMethod = method;
continue;
}
this->methodOut(method, child);
continue;
}
fLineCount = child.fLineCount;
return this->reportError<bool>("method not found");
}
if (Bracket::kSlashSlash == child.fBracket || Bracket::kSlashStar == child.fBracket) {
if (!fDeferComment) {
fDeferComment = &child;
}
fLastComment = &child;
continue;
}
if (MarkType::kMethod == child.fMarkType) {
if (this->internalName(child)) {
continue;
}
const char* bodyEnd = fDeferComment ? fDeferComment->fContentStart - 1 :
fAttrDeprecated ? fAttrDeprecated->fContentStart - 1 :
child.fContentStart;
// FIXME: roll end-trimming into writeBlockTrim call
while (fStart < bodyEnd && ' ' >= bodyEnd[-1]) {
--bodyEnd;
}
int blockSize = (int) (bodyEnd - fStart);
if (blockSize) {
this->writeBlock(blockSize, fStart);
}
startDef = &child;
fStart = child.fContentStart;
methodName = root->fName + "::" + child.fName;
inConstructor = root->fName == child.fName;
fContinuation = child.fContentEnd;
method = root->find(methodName, RootDefinition::AllowParens::kNo);
// if (!method) {
// method = root->find(methodName + "()", RootDefinition::AllowParens::kNo);
// }
if (!method) {
continue;
}
if (method->fCloned) {
clonedMethod = method;
continue;
}
this->methodOut(method, child);
if (fAttrDeprecated) {
startDef = fAttrDeprecated;
fStart = fAttrDeprecated->fContentStart;
fAttrDeprecated = nullptr;
}
continue;
}
if (Definition::Type::kKeyWord == child.fType) {
if (fIndentNext) {
SkDebugf("");
// too soon
#if 0 // makes struct Lattice indent when it oughtn't
if (KeyWord::kEnum == child.fKeyWord) {
fIndent += 4;
}
if (KeyWord::kPublic != child.fKeyWord) {
fIndentNext = false;
}
#endif
}
const Definition* cIncludeStructDef = nullptr;
switch (child.fKeyWord) {
case KeyWord::kStruct:
case KeyWord::kClass:
// if struct contains members, compute their name and comment tabs
if (child.fChildren.size() > 0) {
const ParentPair* testPair = &pair;
while ((testPair = testPair->fPrev)) {
if (KeyWord::kClass == testPair->fParent->fKeyWord) {
inStruct = fInStruct = true;
break;
}
}
}
if (fInStruct) {
// try child; root+child; root->parent+child; etc.
int trial = 0;
const RootDefinition* search = root;
const Definition* parent = search->fParent;
do {
string name;
if (0 == trial) {
name = child.fName;
} else if (1 == trial) {
name = root->fName + "::" + child.fName;
} else {
SkASSERT(parent);
name = parent->fName + "::" + child.fName;
search = parent->asRoot();
parent = search->fParent;
}
fBmhStructDef = search->find(name, RootDefinition::AllowParens::kNo);
} while (!fBmhStructDef && ++trial);
root = const_cast<RootDefinition*>(fBmhStructDef->asRoot());
SkASSERT(root);
fIndent += 4;
this->structSizeMembers(child);
fIndent -= 4;
SkASSERT(!fIndentNext);
fIndentNext = true;
}
if (child.fChildren.size() > 0) {
const char* bodyEnd = fDeferComment ? fDeferComment->fContentStart - 1 :
child.fContentStart;
this->writeBlockTrim((int) (bodyEnd - fStart), fStart);
if (fPendingMethod) {
fIndent -= 4;
fPendingMethod = false;
}
startDef = requireDense ? requireDense : &child;
fStart = requireDense ? requireDense->fContentStart : child.fContentStart;
requireDense = nullptr;
if (!fInStruct && child.fName != root->fName) {
root = &fBmhParser->fClassMap[child.fName];
fRootTopic = root->fParent;
SkASSERT(!root->fVisited);
root->clearVisited();
fIndent = 0;
fBmhStructDef = root;
}
if (child.fName == root->fName) {
if (Definition* parent = root->fParent) {
if (MarkType::kTopic == parent->fMarkType ||
MarkType::kSubtopic == parent->fMarkType) {
const char* commentStart = root->fContentStart;
const char* commentEnd = root->fChildren[0]->fStart;
this->structOut(root, *root, commentStart, commentEnd);
} else {
SkASSERT(0); // incomplete
}
} else {
SkASSERT(0); // incomplete
}
} else {
SkASSERT(fInStruct);
#if 0
fBmhStructDef = root->find(child.fName, RootDefinition::AllowParens::kNo);
if (nullptr == fBmhStructDef) {
fBmhStructDef = root->find(root->fName + "::" + child.fName,
RootDefinition::AllowParens::kNo);
}
if (!fBmhStructDef) {
this->lf(2);
fIndent = 0;
this->writeBlock((int) (fStart - bodyEnd), bodyEnd);
this->lfcr();
continue;
}
#endif
Definition* codeBlock = nullptr;
Definition* nextBlock = nullptr;
for (auto test : fBmhStructDef->fChildren) {
if (MarkType::kCode == test->fMarkType) {
SkASSERT(!codeBlock); // FIXME: check enum for correct order earlier
codeBlock = test;
continue;
}
if (codeBlock) {
nextBlock = test;
break;
}
}
// FIXME: trigger error earlier if inner #Struct or #Class is missing #Code
SkASSERT(nextBlock); // FIXME: check enum for correct order earlier
const char* commentStart = codeBlock->fTerminator;
const char* commentEnd = nextBlock->fStart;
if (fIndentNext) {
// fIndent += 4;
}
fIndentNext = true;
this->structOut(root, *fBmhStructDef, commentStart, commentEnd);
}
fDeferComment = nullptr;
} else {
; // empty forward reference, nothing to do here
}
break;
case KeyWord::kEnum: {
fInEnum = true;
this->enumHeaderOut(root, child);
this->enumSizeItems(child);
} break;
case KeyWord::kConst:
case KeyWord::kConstExpr:
case KeyWord::kStatic:
case KeyWord::kInt:
case KeyWord::kUint8_t:
case KeyWord::kUint16_t:
case KeyWord::kUint32_t:
case KeyWord::kUint64_t:
case KeyWord::kUnsigned:
case KeyWord::kSize_t:
case KeyWord::kFloat:
case KeyWord::kBool:
case KeyWord::kVoid:
if (!memberStart) {
memberStart = &child;
}
break;
case KeyWord::kPublic:
case KeyWord::kPrivate:
case KeyWord::kProtected:
case KeyWord::kFriend:
case KeyWord::kInline:
case KeyWord::kSK_API:
case KeyWord::kTypedef:
break;
case KeyWord::kSK_BEGIN_REQUIRE_DENSE:
requireDense = &child;
break;
default:
SkASSERT(0);
}
if (cIncludeStructDef) {
TextParser structName(&child);
SkAssertResult(structName.skipToEndBracket('{'));
startDef = &child;
fStart = structName.fChar + 1;
this->writeBlock((int) (fStart - child.fStart), child.fStart);
this->lf(2);
fIndent += 4;
if (!this->populate(&child, &pair, const_cast<Definition*>(cIncludeStructDef)->asRoot())) {
return false;
}
// output any remaining definitions at current indent level
const char* structEnd = child.fContentEnd;
SkAssertResult('}' == structEnd[-1]);
--structEnd;
this->writeBlockTrim((int) (structEnd - fStart), fStart);
this->lf(2);
fStart = structEnd;
fIndent -= 4;
fContinuation = nullptr;
fDeferComment = nullptr;
} else if (KeyWord::kUint8_t == child.fKeyWord) {
continue;
} else {
if (fInEnum && KeyWord::kClass == child.fChildren[0]->fKeyWord) {
if (!this->populate(child.fChildren[0], &pair, root)) {
return false;
}
} else {
if (!this->populate(&child, &pair, root)) {
return false;
}
if (KeyWord::kClass == child.fKeyWord || KeyWord::kStruct == child.fKeyWord) {
fStructMemberTab = 0;
if (fInStruct) {
fInStruct = false;
do {
SkASSERT(root);
root = const_cast<RootDefinition*>(root->fParent->asRoot());
} while (MarkType::kTopic == root->fMarkType ||
MarkType::kSubtopic == root->fMarkType);
SkASSERT(MarkType::kStruct == root->fMarkType ||
MarkType::kClass == root->fMarkType);
fPendingMethod = false;
if (startDef) {
fPendingMethod = find_start(startDef, fStart);
}
fOutdentNext = !fPendingMethod;
}
}
}
}
continue;
}
if (Definition::Type::kBracket == child.fType) {
if (KeyWord::kEnum == child.fParent->fKeyWord ||
(KeyWord::kClass == child.fParent->fKeyWord && child.fParent->fParent &&
KeyWord::kEnum == child.fParent->fParent->fKeyWord)) {
SkASSERT(Bracket::kBrace == child.fBracket);
this->enumMembersOut(root, child);
this->writeString("};");
this->lf(2);
startDef = child.fParent;
fStart = child.fParent->fContentEnd;
SkASSERT(';' == fStart[0]);
++fStart;
fDeferComment = nullptr;
fInEnum = false;
if (fIndentNext) {
// fIndent -= 4;
fIndentNext = false;
}
continue;
}
if (fAttrDeprecated) {
continue;
}
fDeferComment = nullptr;
if (KeyWord::kClass == def->fKeyWord || KeyWord::kStruct == def->fKeyWord) {
fIndentNext = true;
}
if (!this->populate(&child, &pair, root)) {
return false;
}
continue;
}
if (Definition::Type::kWord == child.fType) {
if (MarkType::kMember == child.fMarkType) {
if (!memberStart) {
auto iter = def->fTokens.begin();
std::advance(iter, child.fParentIndex - 1);
memberStart = &*iter;
if (!fStructMemberTab) {
SkASSERT(KeyWord::kStruct == def->fParent->fKeyWord);
fIndent += 4;
this->structSizeMembers(*def->fParent);
fIndent -= 4;
// SkASSERT(!fIndentNext);
fIndentNext = true;
}
}
memberEnd = this->structMemberOut(memberStart, child);
startDef = &child;
fStart = child.fContentEnd + 1;
fDeferComment = nullptr;
}
if (child.fMemberStart) {
memberStart = &child;
}
const char attrDeprecated[] = "SK_ATTR_DEPRECATED";
const size_t attrDeprecatedLen = sizeof(attrDeprecated) - 1;
if (attrDeprecatedLen == child.fContentEnd - child.fContentStart &&
!strncmp(attrDeprecated, child.fStart, attrDeprecatedLen)) {
fAttrDeprecated = &child;
}
continue;
}
if (Definition::Type::kPunctuation == child.fType) {
if (Punctuation::kSemicolon == child.fPunctuation) {
memberStart = nullptr;
if (inStruct) {
fInStruct = false;
}
continue;
}
if (Punctuation::kLeftBrace == child.fPunctuation ||
Punctuation::kColon == child.fPunctuation ||
Punctuation::kAsterisk == child.fPunctuation
) {
continue;
}
}
}
return true;
}
bool IncludeWriter::populate(BmhParser& bmhParser) {
bool allPassed = true;
for (auto& includeMapper : fIncludeMap) {
size_t lastSlash = includeMapper.first.rfind('/');
if (string::npos == lastSlash) {
lastSlash = includeMapper.first.rfind('\\');
}
if (string::npos == lastSlash || lastSlash >= includeMapper.first.length() - 1) {
return this->reportError<bool>("malformed include name");
}
string fileName = includeMapper.first.substr(lastSlash + 1);
if (".h" != fileName.substr(fileName.length() - 2)) {
return this->reportError<bool>("expected fileName.h");
}
string skClassName = fileName.substr(0, fileName.length() - 2);
fOut = fopen(fileName.c_str(), "wb");
if (!fOut) {
SkDebugf("could not open output file %s\n", fileName.c_str());
return false;
}
if (bmhParser.fClassMap.end() == bmhParser.fClassMap.find(skClassName)) {
return this->reportError<bool>("could not find bmh class");
}
fBmhParser = &bmhParser;
RootDefinition* root = &bmhParser.fClassMap[skClassName];
fRootTopic = root->fParent;
root->clearVisited();
fStart = includeMapper.second.fContentStart;
fEnd = includeMapper.second.fContentEnd;
allPassed &= this->populate(&includeMapper.second, nullptr, root);
this->writeBlock((int) (fEnd - fStart), fStart);
fIndent = 0;
this->lfcr();
this->writePending();
fclose(fOut);
SkDebugf("wrote %s\n", fileName.c_str());
}
return allPassed;
}
// change Xxx_Xxx to xxx xxx
static string ConvertRef(const string str, bool first) {
string substitute;
for (char c : str) {
if ('_' == c) {
c = ' '; // change Xxx_Xxx to xxx xxx
} else if (isupper(c) && !first) {
c = tolower(c);
}
substitute += c;
first = false;
}
return substitute;
}
string IncludeWriter::resolveMethod(const char* start, const char* end, bool first) {
string methodname(start, end - start);
if (string::npos != methodname.find("()")) {
return "";
}
string substitute;
auto rootDefIter = fBmhParser->fMethodMap.find(methodname);
if (fBmhParser->fMethodMap.end() != rootDefIter) {
substitute = methodname + "()";
} else {
RootDefinition* parent = nullptr;
for (auto candidate : fRootTopic->fChildren) {
if (MarkType::kClass == candidate->fMarkType
|| MarkType::kStruct == candidate->fMarkType) {
parent = candidate->asRoot();
break;
}
}
SkASSERT(parent);
auto defRef = parent->find(parent->fName + "::" + methodname,
RootDefinition::AllowParens::kNo);
if (defRef && MarkType::kMethod == defRef->fMarkType) {
substitute = methodname + "()";
}
}
if (fMethodDef && methodname == fMethodDef->fName) {
TextParser report(fBmhMethod);
report.reportError("method should not include references to itself");
return "";
}
if (fBmhMethod) {
for (auto child : fBmhMethod->fChildren) {
if (MarkType::kParam != child->fMarkType) {
continue;
}
if (methodname == child->fName) {
return "";
}
}
}
return substitute;
}
string IncludeWriter::resolveRef(const char* start, const char* end, bool first,
RefType* refType) {
// look up Xxx_Xxx
string undername(start, end - start);
for (const auto& external : fBmhParser->fExternals) {
if (external.fName == undername) {
*refType = RefType::kExternal;
return external.fName;
}
}
*refType = RefType::kNormal;
SkASSERT(string::npos == undername.find(' '));
const Definition* rootDef = nullptr;
{
auto rootDefIter = fBmhParser->fTopicMap.find(undername);
if (fBmhParser->fTopicMap.end() != rootDefIter) {
rootDef = rootDefIter->second;
} else {
string prefixedName = fRootTopic->fName + '_' + undername;
rootDefIter = fBmhParser->fTopicMap.find(prefixedName);
if (fBmhParser->fTopicMap.end() != rootDefIter) {
rootDef = rootDefIter->second;
} else if (fBmhStructDef) {
string localPrefix = fBmhStructDef->fFiddle + '_' + undername;
rootDefIter = fBmhParser->fTopicMap.find(localPrefix);
if (fBmhParser->fTopicMap.end() != rootDefIter) {
rootDef = rootDefIter->second;
}
} else {
auto aliasIter = fBmhParser->fAliasMap.find(undername);
if (fBmhParser->fAliasMap.end() != aliasIter) {
rootDef = aliasIter->second->fParent;
} else if (!first) {
SkDebugf("unfound: %s\n", undername.c_str());
this->reportError("reference unfound");
return "";
}
}
}
}
string substitute;
if (rootDef) {
for (auto child : rootDef->fChildren) {
if (MarkType::kSubstitute == child->fMarkType) {
substitute = string(child->fContentStart,
(int) (child->fContentEnd - child->fContentStart));
break;
}
}
if (!substitute.length()) {
for (auto child : rootDef->fChildren) {
if (MarkType::kClass == child->fMarkType ||
MarkType::kStruct == child->fMarkType ||
(MarkType::kEnum == child->fMarkType && !child->fAnonymous) ||
MarkType::kEnumClass == child->fMarkType) {
substitute = child->fName;
if (MarkType::kEnum == child->fMarkType && fInEnum) {
size_t parentClassEnd = substitute.find("::");
SkASSERT(string::npos != parentClassEnd);
substitute = substitute.substr(parentClassEnd + 2);
}
break;
}
}
}
if (!substitute.length()) {
auto parent = rootDef->fParent;
if (parent) {
if (MarkType::kClass == parent->fMarkType ||
MarkType::kStruct == parent->fMarkType ||
(MarkType::kEnum == parent->fMarkType && !parent->fAnonymous) ||
MarkType::kEnumClass == parent->fMarkType) {
if (parent->fParent != fRootTopic) {
substitute = parent->fName;
size_t under = undername.find('_');
SkASSERT(string::npos != under);
string secondHalf(&undername[under], (size_t) (undername.length() - under));
substitute += ConvertRef(secondHalf, false);
} else {
substitute += ConvertRef(undername, first);
}
}
}
}
}
// Ensure first word after period is capitalized if substitute is lower cased.
if (first && isupper(start[0]) && substitute.length() > 0 && islower(substitute[0])) {
substitute[0] = start[0];
}
return substitute;
}
int IncludeWriter::lookupMethod(const PunctuationState punctuation, const Word word,
const int lastSpace, const int run, int lastWrite, const char* data,
bool hasIndirection) {
int wordStart = lastSpace;
while (' ' >= data[wordStart]) {
++wordStart;
}
const int wordEnd = PunctuationState::kDelimiter == punctuation ||
PunctuationState::kPeriod == punctuation ? run - 1 : run;
string temp;
if (hasIndirection && '(' != data[wordEnd - 1] && ')' != data[wordEnd - 1]) {
// FIXME: hard-coded to assume a.b or a->b is a.b() or a->b().
// need to check class a for member b to see if this is so
TextParser parser(fFileName, &data[wordStart], &data[wordEnd], fLineCount);
const char* indirection = parser.anyOf(".>");
if (&data[wordEnd] <= &indirection[2] || 'f' != indirection[1] ||
!isupper(indirection[2])) {
temp = string(&data[wordStart], wordEnd - wordStart) + "()";
}
} else {
temp = this->resolveMethod(&data[wordStart], &data[wordEnd], Word::kFirst == word);
}
if (temp.length()) {
if (wordStart > lastWrite) {
SkASSERT(data[wordStart - 1] >= ' ');
if (' ' == data[lastWrite]) {
this->writeSpace();
}
this->writeBlockTrim(wordStart - lastWrite, &data[lastWrite]);
if (' ' == data[wordStart - 1]) {
this->writeSpace();
}
}
SkASSERT(temp[temp.length() - 1] > ' ');
this->writeString(temp.c_str());
lastWrite = wordEnd;
}
return lastWrite;
}
int IncludeWriter::lookupReference(const PunctuationState punctuation, const Word word,
const int start, const int run, int lastWrite, const char last, const char* data) {
const int end = PunctuationState::kDelimiter == punctuation ||
PunctuationState::kPeriod == punctuation ? run - 1 : run;
RefType refType = RefType::kUndefined;
string resolved = string(&data[start], (size_t) (end - start));
string temp = this->resolveRef(&data[start], &data[end], Word::kFirst == word, &refType);
if (!temp.length()) {
if (Word::kFirst != word && '_' != last) {
temp = ConvertRef(resolved, false);
}
}
if (temp.length()) {
if (start > lastWrite) {
SkASSERT(data[start - 1] >= ' ');
if (' ' == data[lastWrite]) {
this->writeSpace();
}
this->writeBlockTrim(start - lastWrite, &data[lastWrite]);
if (' ' == data[start - 1]) {
this->writeSpace();
}
}
SkASSERT(temp[temp.length() - 1] > ' ');
this->writeString(temp.c_str());
lastWrite = end;
}
return lastWrite;
}
/* returns true if rewriteBlock wrote linefeeds */
IncludeWriter::Wrote IncludeWriter::rewriteBlock(int size, const char* data, Phrase phrase) {
bool wroteLineFeeds = false;
while (size > 0 && data[0] <= ' ') {
--size;
++data;
}
while (size > 0 && data[size - 1] <= ' ') {
--size;
}
if (0 == size) {
return Wrote::kNone;
}
int run = 0;
Word word = Word::kStart;
PunctuationState punctuation = Phrase::kNo == phrase ?
PunctuationState::kStart : PunctuationState::kSpace;
int start = 0;
int lastWrite = 0;
int lineFeeds = 0;
int lastPrintable = 0;
int lastSpace = -1;
char c = 0;
char last;
bool embeddedIndirection = false;
bool embeddedSymbol = false;
bool hasLower = false;
bool hasUpper = false;
bool hasIndirection = false;
bool hasSymbol = false;
while (run < size) {
last = c;
c = data[run];
SkASSERT(' ' <= c || '\n' == c);
if (lineFeeds && ' ' < c) {
if (lastPrintable >= lastWrite) {
if (' ' == data[lastWrite]) {
this->writeSpace();
}
this->writeBlock(lastPrintable - lastWrite + 1, &data[lastWrite]);
}
if (lineFeeds > 1) {
this->lf(2);
}
this->lfcr(); // defer the indent until non-whitespace is seen
lastWrite = run;
lineFeeds = 0;
}
if (' ' < c) {
lastPrintable = run;
}
switch (c) {
case '\n':
++lineFeeds;
wroteLineFeeds = true;
case ' ':
switch (word) {
case Word::kStart:
break;
case Word::kUnderline:
case Word::kCap:
case Word::kFirst:
if (!hasLower) {
break;
}
lastWrite = this->lookupReference(punctuation, word, start, run,
lastWrite, last, data);
break;
case Word::kMixed:
if (hasUpper && hasLower && !hasSymbol && lastSpace > 0) {
lastWrite = this->lookupMethod(punctuation, word, lastSpace, run,
lastWrite, data, hasIndirection && !hasSymbol);
}
break;
default:
SkASSERT(0);
}
punctuation = PunctuationState::kPeriod == punctuation ||
(PunctuationState::kStart == punctuation && ' ' >= last) ?
PunctuationState::kStart : PunctuationState::kSpace;
word = Word::kStart;
embeddedIndirection = false;
embeddedSymbol = false;
hasLower = false;
hasUpper = false;
hasIndirection = false;
hasSymbol = false;
lastSpace = run;
break;
case '.':
switch (word) {
case Word::kStart:
punctuation = PunctuationState::kDelimiter;
case Word::kCap:
case Word::kFirst:
case Word::kUnderline:
case Word::kMixed:
if (PunctuationState::kDelimiter == punctuation ||
PunctuationState::kPeriod == punctuation) {
word = Word::kMixed;
}
punctuation = PunctuationState::kPeriod;
break;
default:
SkASSERT(0);
}
embeddedIndirection = true;
break;
case ',': case ';': case ':':
switch (word) {
case Word::kStart:
punctuation = PunctuationState::kDelimiter;
case Word::kCap:
case Word::kFirst:
case Word::kUnderline:
case Word::kMixed:
if (PunctuationState::kDelimiter == punctuation ||
PunctuationState::kPeriod == punctuation) {
word = Word::kMixed;
}
punctuation = PunctuationState::kDelimiter;
break;
default:
SkASSERT(0);
}
embeddedSymbol = true;
break;
case '>':
if ('-' == last) {
embeddedIndirection = true;
break;
}
case '\'': // possessive apostrophe isn't treated as delimiting punctation
case '\"': // quote is passed straight through
case '=':
case '!': // assumed not to be punctuation, but a programming symbol
case '&': case '<': case '{': case '}': case '/': case '*': case '[': case ']':
word = Word::kMixed;
embeddedSymbol = true;
break;
case '(':
if (' ' == last) {
punctuation = PunctuationState::kDelimiter;
} else {
word = Word::kMixed;
}
embeddedSymbol = true;
break;
case ')': // assume word type has already been set
punctuation = PunctuationState::kDelimiter;
embeddedSymbol = true;
break;
case '_':
switch (word) {
case Word::kStart:
word = Word::kMixed;
break;
case Word::kCap:
case Word::kFirst:
case Word::kUnderline:
word = Word::kUnderline;
break;
case Word::kMixed:
break;
default:
SkASSERT(0);
}
hasSymbol |= embeddedSymbol;
break;
case '+':
// hackery to allow C++
SkASSERT('C' == last || '+' == last); // FIXME: don't allow + outside of #Formula
break;
case 'A': case 'B': case 'C': case 'D': case 'E':
case 'F': case 'G': case 'H': case 'I': case 'J':
case 'K': case 'L': case 'M': case 'N': case 'O':
case 'P': case 'Q': case 'R': case 'S': case 'T':
case 'U': case 'V': case 'W': case 'X': case 'Y':
case 'Z':
switch (word) {
case Word::kStart:
word = PunctuationState::kStart == punctuation ? Word::kFirst : Word::kCap;
start = run;
break;
case Word::kCap:
case Word::kFirst:
if (!isupper(last) && '~' != last) {
word = Word::kMixed;
}
break;
case Word::kUnderline:
// some word in Xxx_XXX_Xxx can be all upper, but all can't: XXX_XXX
if ('_' != last && !isupper(last)) {
word = Word::kMixed;
}
break;
case Word::kMixed:
break;
default:
SkASSERT(0);
}
hasUpper = true;
if (PunctuationState::kPeriod == punctuation ||
PunctuationState::kDelimiter == punctuation) {
word = Word::kMixed;
}
hasIndirection |= embeddedIndirection;
hasSymbol |= embeddedSymbol;
break;
case 'a': case 'b': case 'c': case 'd': case 'e':
case 'f': case 'g': case 'h': case 'i': case 'j':
case 'k': case 'l': case 'm': case 'n': case 'o':
case 'p': case 'q': case 'r': case 's': case 't':
case 'u': case 'v': case 'w': case 'x': case 'y':
case 'z':
case '0': case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
case '-':
switch (word) {
case Word::kStart:
word = Word::kMixed;
break;
case Word::kMixed:
case Word::kCap:
case Word::kFirst:
case Word::kUnderline:
break;
default:
SkASSERT(0);
}
hasLower = true;
punctuation = PunctuationState::kStart;
hasIndirection |= embeddedIndirection;
hasSymbol |= embeddedSymbol;
break;
case '~':
SkASSERT(Word::kStart == word);
word = PunctuationState::kStart == punctuation ? Word::kFirst : Word::kCap;
start = run;
hasUpper = true;
hasIndirection |= embeddedIndirection;
hasSymbol |= embeddedSymbol;
break;
default:
SkASSERT(0);
}
++run;
}
if ((word == Word::kCap || word == Word::kFirst || word == Word::kUnderline) && hasLower) {
lastWrite = this->lookupReference(punctuation, word, start, run, lastWrite, last, data);
} else if (word == Word::kMixed && hasUpper && hasLower && !hasSymbol && lastSpace > 0) {
lastWrite = this->lookupMethod(punctuation, word, lastSpace, run, lastWrite, data,
hasIndirection && !hasSymbol);
}
if (run > lastWrite) {
if (' ' == data[lastWrite]) {
this->writeSpace();
}
this->writeBlock(run - lastWrite, &data[lastWrite]);
}
return wroteLineFeeds ? Wrote::kLF : Wrote::kChars;
}