bsc/vendor/github.com/robertkrimen/otto/ast/comments.go
Péter Szilágyi 289b30715d Godeps, vendor: convert dependency management to trash (#3198)
This commit converts the dependency management from Godeps to the vendor
folder, also switching the tool from godep to trash. Since the upstream tool
lacks a few features proposed via a few PRs, until those PRs are merged in
(if), use github.com/karalabe/trash.

You can update dependencies via trash --update.

All dependencies have been updated to their latest version.

Parts of the build system are reworked to drop old notions of Godeps and
invocation of the go vet command so that it doesn't run against the vendor
folder, as that will just blow up during vetting.

The conversion drops OpenCL (and hence GPU mining support) from ethash and our
codebase. The short reasoning is that there's noone to maintain and having
opencl libs in our deps messes up builds as go install ./... tries to build
them, failing with unsatisfied link errors for the C OpenCL deps.

golang.org/x/net/context is not vendored in. We expect it to be fetched by the
user (i.e. using go get). To keep ci.go builds reproducible the package is
"vendored" in build/_vendor.
2016-10-28 19:05:01 +02:00

279 lines
7.1 KiB
Go

package ast
import (
"fmt"
"github.com/robertkrimen/otto/file"
)
// CommentPosition determines where the comment is in a given context
type CommentPosition int
const (
_ CommentPosition = iota
LEADING // Before the pertinent expression
TRAILING // After the pertinent expression
KEY // Before a key in an object
COLON // After a colon in a field declaration
FINAL // Final comments in a block, not belonging to a specific expression or the comment after a trailing , in an array or object literal
IF // After an if keyword
WHILE // After a while keyword
DO // After do keyword
FOR // After a for keyword
WITH // After a with keyword
TBD
)
// Comment contains the data of the comment
type Comment struct {
Begin file.Idx
Text string
Position CommentPosition
}
// NewComment creates a new comment
func NewComment(text string, idx file.Idx) *Comment {
comment := &Comment{
Begin: idx,
Text: text,
Position: TBD,
}
return comment
}
// String returns a stringified version of the position
func (cp CommentPosition) String() string {
switch cp {
case LEADING:
return "Leading"
case TRAILING:
return "Trailing"
case KEY:
return "Key"
case COLON:
return "Colon"
case FINAL:
return "Final"
case IF:
return "If"
case WHILE:
return "While"
case DO:
return "Do"
case FOR:
return "For"
case WITH:
return "With"
default:
return "???"
}
}
// String returns a stringified version of the comment
func (c Comment) String() string {
return fmt.Sprintf("Comment: %v", c.Text)
}
// Comments defines the current view of comments from the parser
type Comments struct {
// CommentMap is a reference to the parser comment map
CommentMap CommentMap
// Comments lists the comments scanned, not linked to a node yet
Comments []*Comment
// future lists the comments after a line break during a sequence of comments
future []*Comment
// Current is node for which comments are linked to
Current Expression
// wasLineBreak determines if a line break occured while scanning for comments
wasLineBreak bool
// primary determines whether or not processing a primary expression
primary bool
// afterBlock determines whether or not being after a block statement
afterBlock bool
}
func NewComments() *Comments {
comments := &Comments{
CommentMap: CommentMap{},
}
return comments
}
func (c *Comments) String() string {
return fmt.Sprintf("NODE: %v, Comments: %v, Future: %v(LINEBREAK:%v)", c.Current, len(c.Comments), len(c.future), c.wasLineBreak)
}
// FetchAll returns all the currently scanned comments,
// including those from the next line
func (c *Comments) FetchAll() []*Comment {
defer func() {
c.Comments = nil
c.future = nil
}()
return append(c.Comments, c.future...)
}
// Fetch returns all the currently scanned comments
func (c *Comments) Fetch() []*Comment {
defer func() {
c.Comments = nil
}()
return c.Comments
}
// ResetLineBreak marks the beginning of a new statement
func (c *Comments) ResetLineBreak() {
c.wasLineBreak = false
}
// MarkPrimary will mark the context as processing a primary expression
func (c *Comments) MarkPrimary() {
c.primary = true
c.wasLineBreak = false
}
// AfterBlock will mark the context as being after a block.
func (c *Comments) AfterBlock() {
c.afterBlock = true
}
// AddComment adds a comment to the view.
// Depending on the context, comments are added normally or as post line break.
func (c *Comments) AddComment(comment *Comment) {
if c.primary {
if !c.wasLineBreak {
c.Comments = append(c.Comments, comment)
} else {
c.future = append(c.future, comment)
}
} else {
if !c.wasLineBreak || (c.Current == nil && !c.afterBlock) {
c.Comments = append(c.Comments, comment)
} else {
c.future = append(c.future, comment)
}
}
}
// MarkComments will mark the found comments as the given position.
func (c *Comments) MarkComments(position CommentPosition) {
for _, comment := range c.Comments {
if comment.Position == TBD {
comment.Position = position
}
}
for _, c := range c.future {
if c.Position == TBD {
c.Position = position
}
}
}
// Unset the current node and apply the comments to the current expression.
// Resets context variables.
func (c *Comments) Unset() {
if c.Current != nil {
c.applyComments(c.Current, c.Current, TRAILING)
c.Current = nil
}
c.wasLineBreak = false
c.primary = false
c.afterBlock = false
}
// SetExpression sets the current expression.
// It is applied the found comments, unless the previous expression has not been unset.
// It is skipped if the node is already set or if it is a part of the previous node.
func (c *Comments) SetExpression(node Expression) {
// Skipping same node
if c.Current == node {
return
}
if c.Current != nil && c.Current.Idx1() == node.Idx1() {
c.Current = node
return
}
previous := c.Current
c.Current = node
// Apply the found comments and futures to the node and the previous.
c.applyComments(node, previous, TRAILING)
}
// PostProcessNode applies all found comments to the given node
func (c *Comments) PostProcessNode(node Node) {
c.applyComments(node, nil, TRAILING)
}
// applyComments applies both the comments and the future comments to the given node and the previous one,
// based on the context.
func (c *Comments) applyComments(node, previous Node, position CommentPosition) {
if previous != nil {
c.CommentMap.AddComments(previous, c.Comments, position)
c.Comments = nil
} else {
c.CommentMap.AddComments(node, c.Comments, position)
c.Comments = nil
}
// Only apply the future comments to the node if the previous is set.
// This is for detecting end of line comments and which node comments on the following lines belongs to
if previous != nil {
c.CommentMap.AddComments(node, c.future, position)
c.future = nil
}
}
// AtLineBreak will mark a line break
func (c *Comments) AtLineBreak() {
c.wasLineBreak = true
}
// CommentMap is the data structure where all found comments are stored
type CommentMap map[Node][]*Comment
// AddComment adds a single comment to the map
func (cm CommentMap) AddComment(node Node, comment *Comment) {
list := cm[node]
list = append(list, comment)
cm[node] = list
}
// AddComments adds a slice of comments, given a node and an updated position
func (cm CommentMap) AddComments(node Node, comments []*Comment, position CommentPosition) {
for _, comment := range comments {
if comment.Position == TBD {
comment.Position = position
}
cm.AddComment(node, comment)
}
}
// Size returns the size of the map
func (cm CommentMap) Size() int {
size := 0
for _, comments := range cm {
size += len(comments)
}
return size
}
// MoveComments moves comments with a given position from a node to another
func (cm CommentMap) MoveComments(from, to Node, position CommentPosition) {
for i, c := range cm[from] {
if c.Position == position {
cm.AddComment(to, c)
// Remove the comment from the "from" slice
cm[from][i] = cm[from][len(cm[from])-1]
cm[from][len(cm[from])-1] = nil
cm[from] = cm[from][:len(cm[from])-1]
}
}
}