// writeString writes the string s to p.output and updates p.pos, p.out,
// and p.last. If isLit is set, s is escaped w/ tabwriter.Escape characters
// to protect s from being interpreted by the tabwriter.
//
// Note: writeString is only used to write Go tokens, literals, and
// comments, all of which must be written literally. Thus, it is correct
// to always set isLit = true. However, setting it explicitly only when
// needed (i.e., when we don't know that s contains no tabs or line breaks)
// avoids processing extra escape characters and reduces run time of the
// printer benchmark by up to 10%.
//
func (p *printer) writeString(pos token.Position, s string, isLit bool) {
if p.out.Column == 1 {
p.atLineBegin(pos)
}
if pos.IsValid() {
// update p.pos (if pos is invalid, continue with existing p.pos)
// Note: Must do this after handling line beginnings because
// atLineBegin updates p.pos if there's indentation, but p.pos
// is the position of s.
p.pos = pos
}
if isLit {
// Protect s such that is passes through the tabwriter
// unchanged. Note that valid Go programs cannot contain
// tabwriter.Escape bytes since they do not appear in legal
// UTF-8 sequences.
p.output = append(p.output, tabwriter.Escape)
}
if debug {
p.output = append(p.output, fmt.Sprintf("/*%s*/", pos)...) // do not update p.pos!
}
p.output = append(p.output, s...)
// update positions
nlines := 0
var li int // index of last newline; valid if nlines > 0
for i := 0; i < len(s); i++ {
// Go tokens cannot contain '\f' - no need to look for it
if s[i] == '\n' {
nlines++
li = i
}
}
p.pos.Offset += len(s)
if nlines > 0 {
p.pos.Line += nlines
p.out.Line += nlines
c := len(s) - li
p.pos.Column = c
p.out.Column = c
} else {
p.pos.Column += len(s)
p.out.Column += len(s)
}
if isLit {
p.output = append(p.output, tabwriter.Escape)
}
p.last = p.pos
}
// atLineBegin emits a //line comment if necessary and prints indentation.
func (p *printer) atLineBegin(pos token.Position) {
// write a //line comment if necessary
if p.Config.Mode&SourcePos != 0 && pos.IsValid() && (p.out.Line != pos.Line || p.out.Filename != pos.Filename) {
p.output = append(p.output, tabwriter.Escape) // protect '\n' in //line from tabwriter interpretation
p.output = append(p.output, fmt.Sprintf("//line %s:%d\n", pos.Filename, pos.Line)...)
p.output = append(p.output, tabwriter.Escape)
// p.out must match the //line comment
p.out.Filename = pos.Filename
p.out.Line = pos.Line
}
// write indentation
// use "hard" htabs - indentation columns
// must not be discarded by the tabwriter
n := p.Config.Indent + p.indent // include base indentation
for i := 0; i < n; i++ {
p.output = append(p.output, '\t')
}
// update positions
p.pos.Offset += n
p.pos.Column += n
p.out.Column += n
}
// writeCommentPrefix writes the whitespace before a comment.
// If there is any pending whitespace, it consumes as much of
// it as is likely to help position the comment nicely.
// pos is the comment position, next the position of the item
// after all pending comments, prev is the previous comment in
// a group of comments (or nil), and tok is the next token.
//
func (p *printer) writeCommentPrefix(pos, next token.Position, prev, comment *ast.Comment, tok token.Token) {
if len(p.output) == 0 {
// the comment is the first item to be printed - don't write any whitespace
return
}
if pos.IsValid() && pos.Filename != p.last.Filename {
// comment in a different file - separate with newlines
p.writeByte('\f', maxNewlines)
return
}
if pos.Line == p.last.Line && (prev == nil || prev.Text[1] != '/') {
// comment on the same line as last item:
// separate with at least one separator
hasSep := false
if prev == nil {
// first comment of a comment group
j := 0
for i, ch := range p.wsbuf {
switch ch {
case blank:
// ignore any blanks before a comment
p.wsbuf[i] = ignore
continue
case vtab:
// respect existing tabs - important
// for proper formatting of commented structs
hasSep = true
continue
case indent:
// apply pending indentation
continue
}
j = i
break
}
p.writeWhitespace(j)
}
// make sure there is at least one separator
if !hasSep {
sep := byte('\t')
if pos.Line == next.Line {
// next item is on the same line as the comment
// (which must be a /*-style comment): separate
// with a blank instead of a tab
sep = ' '
}
p.writeByte(sep, 1)
}
} else {
// comment on a different line:
// separate with at least one line break
droppedLinebreak := false
j := 0
for i, ch := range p.wsbuf {
switch ch {
case blank, vtab:
// ignore any horizontal whitespace before line breaks
p.wsbuf[i] = ignore
continue
case indent:
// apply pending indentation
continue
case unindent:
// if this is not the last unindent, apply it
// as it is (likely) belonging to the last
// construct (e.g., a multi-line expression list)
// and is not part of closing a block
if i+1 < len(p.wsbuf) && p.wsbuf[i+1] == unindent {
continue
}
// if the next token is not a closing }, apply the unindent
// if it appears that the comment is aligned with the
// token; otherwise assume the unindent is part of a
// closing block and stop (this scenario appears with
// comments before a case label where the comments
// apply to the next case instead of the current one)
if tok != token.RBRACE && pos.Column == next.Column {
continue
}
case newline, formfeed:
p.wsbuf[i] = ignore
droppedLinebreak = prev == nil // record only if first comment of a group
}
j = i
break
}
p.writeWhitespace(j)
// determine number of linebreaks before the comment
n := 0
if pos.IsValid() && p.last.IsValid() {
n = pos.Line - p.last.Line
if n < 0 { // should never happen
n = 0
}
}
// at the package scope level only (p.indent == 0),
// add an extra newline if we dropped one before:
// this preserves a blank line before documentation
// comments at the package scope level (issue 2570)
if p.indent == 0 && droppedLinebreak {
n++
}
// make sure there is at least one line break
// if the previous comment was a line comment
if n == 0 && prev != nil && prev.Text[1] == '/' {
n = 1
}
if n > 0 {
// use formfeeds to break columns before a comment;
// this is analogous to using formfeeds to separate
// individual lines of /*-style comments
p.writeByte('\f', nlimit(n))
}
}
}
// NewCommentMap creates a new comment map by associating comment groups
// of the comments list with the nodes of the AST specified by node.
//
// A comment group g is associated with a node n if:
//
// - g starts on the same line as n ends
// - g starts on the line immediately following n, and there is
// at least one empty line after g and before the next node
// - g starts before n and is not associated to the node before n
// via the previous rules
//
// NewCommentMap tries to associate a comment group to the "largest"
// node possible: For instance, if the comment is a line comment
// trailing an assignment, the comment is associated with the entire
// assignment rather than just the last operand in the assignment.
//
func NewCommentMap(fset *token.FileSet, node Node, comments []*CommentGroup) CommentMap {
if len(comments) == 0 {
return nil // no comments to map
}
cmap := make(CommentMap)
// set up comment reader r
tmp := make([]*CommentGroup, len(comments))
copy(tmp, comments) // don't change incoming comments
sortComments(tmp)
r := commentListReader{fset: fset, list: tmp} // !r.eol() because len(comments) > 0
r.next()
// create node list in lexical order
nodes := nodeList(node)
nodes = append(nodes, nil) // append sentinel
// set up iteration variables
var (
p Node // previous node
pend token.Position // end of p
pg Node // previous node group (enclosing nodes of "importance")
pgend token.Position // end of pg
stack nodeStack // stack of node groups
)
for _, q := range nodes {
var qpos token.Position
if q != nil {
qpos = fset.Position(q.Pos()) // current node position
} else {
// set fake sentinel position to infinity so that
// all comments get processed before the sentinel
const infinity = 1 << 30
qpos.Offset = infinity
qpos.Line = infinity
}
// process comments before current node
for r.end.Offset <= qpos.Offset {
// determine recent node group
if top := stack.pop(r.comment.Pos()); top != nil {
pg = top
pgend = fset.Position(pg.End())
}
// Try to associate a comment first with a node group
// (i.e., a node of "importance" such as a declaration);
// if that fails, try to associate it with the most recent
// node.
// TODO(gri) try to simplify the logic below
var assoc Node
switch {
case pg != nil &&
(pgend.Line == r.pos.Line ||
pgend.Line+1 == r.pos.Line && r.end.Line+1 < qpos.Line):
// 1) comment starts on same line as previous node group ends, or
// 2) comment starts on the line immediately after the
// previous node group and there is an empty line before
// the current node
// => associate comment with previous node group
assoc = pg
case p != nil &&
(pend.Line == r.pos.Line ||
pend.Line+1 == r.pos.Line && r.end.Line+1 < qpos.Line ||
q == nil):
// same rules apply as above for p rather than pg,
// but also associate with p if we are at the end (q == nil)
assoc = p
default:
// otherwise, associate comment with current node
if q == nil {
// we can only reach here if there was no p
// which would imply that there were no nodes
panic("internal error: no comments should be associated with sentinel")
}
assoc = q
}
cmap.addComment(assoc, r.comment)
if r.eol() {
return cmap
}
r.next()
}
// update previous node
p = q
pend = fset.Position(p.End())
// update previous node group if we see an "important" node
switch q.(type) {
case *File, *Field, Decl, Spec, Stmt:
stack.push(q)
}
}
return cmap
}
// Verify that calling Scan() provides the correct results.
func TestScan(t *testing.T) {
whitespace_linecount := newlineCount(whitespace)
// error handler
eh := func(_ token.Position, msg string) {
t.Errorf("error handler called (msg = %s)", msg)
}
// verify scan
var s Scanner
s.Init(fset.AddFile("", fset.Base(), len(source)), source, eh, ScanComments|dontInsertSemis)
// set up expected position
epos := token.Position{
Filename: "",
Offset: 0,
Line: 1,
Column: 1,
}
index := 0
for {
pos, tok, lit := s.Scan()
// check position
if tok == token.EOF {
// correction for EOF
epos.Line = newlineCount(string(source))
epos.Column = 2
}
checkPos(t, lit, pos, epos)
// check token
e := elt{token.EOF, "", special}
if index < len(tokens) {
e = tokens[index]
index++
}
if tok != e.tok {
t.Errorf("bad token for %q: got %s, expected %s", lit, tok, e.tok)
}
// check token class
if tokenclass(tok) != e.class {
t.Errorf("bad class for %q: got %d, expected %d", lit, tokenclass(tok), e.class)
}
// check literal
elit := ""
switch e.tok {
case token.COMMENT:
// no CRs in comments
elit = string(stripCR([]byte(e.lit)))
//-style comment literal doesn't contain newline
if elit[1] == '/' {
elit = elit[0 : len(elit)-1]
}
case token.IDENT:
elit = e.lit
case token.SEMICOLON:
elit = ";"
default:
if e.tok.IsLiteral() {
// no CRs in raw string literals
elit = e.lit
if elit[0] == '`' {
elit = string(stripCR([]byte(elit)))
}
} else if e.tok.IsKeyword() {
elit = e.lit
}
}
if lit != elit {
t.Errorf("bad literal for %q: got %q, expected %q", lit, lit, elit)
}
if tok == token.EOF {
break
}
// update position
epos.Offset += len(e.lit) + len(whitespace)
epos.Line += newlineCount(e.lit) + whitespace_linecount
}
if s.ErrorCount != 0 {
t.Errorf("found %d errors", s.ErrorCount)
}
}