v/vlib/toml/parser/parser.v

933 lines
26 KiB
V
Raw Normal View History

2021-09-24 20:13:52 +02:00
// Copyright (c) 2021 Lars Pontoppidan. All rights reserved.
// Use of this source code is governed by an MIT license
// that can be found in the LICENSE file.
module parser
import toml.ast
import toml.checker
import toml.util
import toml.token
import toml.scanner
2021-09-26 06:34:47 +02:00
// Parser contains the necessary fields for keeping the state of the parsing process.
2021-09-24 20:13:52 +02:00
pub struct Parser {
pub:
config Config
mut:
scanner &scanner.Scanner
prev_tok token.Token
tok token.Token
peek_tok token.Token
tokens []token.Token // To be able to peek more than one token ahead.
2021-09-24 20:13:52 +02:00
skip_next bool
// The root map (map is called table in TOML world)
root_map map[string]ast.Value
2021-09-24 20:13:52 +02:00
root_map_key string
// Array of Tables state
last_aot string
last_aot_index int
// Root of the tree
ast_root &ast.Root = &ast.Root{}
}
2021-09-26 06:34:47 +02:00
// Config is used to configure a Parser instance.
// `run_checks` is used to en- or disable running of the strict `checker.Checker` type checks.
2021-09-24 20:13:52 +02:00
pub struct Config {
pub:
scanner &scanner.Scanner
run_checks bool = true
}
// new_parser returns a new, stack allocated, `Parser`.
pub fn new_parser(config Config) Parser {
return Parser{
config: config
scanner: config.scanner
}
}
// init initializes the parser.
pub fn (mut p Parser) init() ? {
p.root_map = map[string]ast.Value{}
p.tokens << p.scanner.scan() ?
2021-09-24 20:13:52 +02:00
p.next() ?
}
// run_checker validates the parsed `ast.Value` nodes in the
2021-09-24 20:13:52 +02:00
// the generated AST.
fn (mut p Parser) run_checker() ? {
if p.config.run_checks {
chckr := checker.Checker{
scanner: p.scanner
}
chckr.check(p.root_map) ?
2021-10-26 15:58:05 +02:00
for comment in p.ast_root.comments {
chckr.check_comment(comment) ?
}
2021-09-24 20:13:52 +02:00
}
}
// parse starts parsing the input and returns the root
// of the generated AST.
pub fn (mut p Parser) parse() ?&ast.Root {
p.init() ?
p.root_table() ?
p.run_checker() ?
p.ast_root.table = p.root_map
return p.ast_root
}
// next forwards the parser to the next token.
fn (mut p Parser) next() ? {
p.prev_tok = p.tok
p.tok = p.peek_tok
if p.tokens.len > 0 {
p.peek_tok = p.tokens.pop()
p.peek(1) ?
} else {
p.peek(1) ?
p.peek_tok = p.tokens.pop()
}
}
// peek peeks forward `n` tokens.
// peek returns `.unknown` if it can not peek ahead long enough.
fn (mut p Parser) peek(n int) ?token.Token {
if n < 0 {
return error(@MOD + '.' + @STRUCT + '.' + @FN + ' peeking backwards is not supported.')
}
if n == 0 {
return p.peek_tok
} else {
// n >= 1
if n <= p.tokens.len {
return p.tokens[n - 1]
} else {
mut token := token.Token{}
mut count := n - p.tokens.len
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'buffering $count tokens...')
for token.kind != .eof && count != 0 {
token = p.scanner.scan() ?
p.tokens << token
count--
}
return token
}
}
2021-09-24 20:13:52 +02:00
}
// check forwards the parser to the next token if the current
// token's `Kind` is equal that of `check_token`.
2021-09-24 20:13:52 +02:00
fn (mut p Parser) check(check_token token.Kind) ? {
if p.tok.kind == check_token {
p.next() ?
} else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' expected token "$check_token" but found "$p.tok.kind" in this (excerpt): "...${p.excerpt()}..."')
}
}
// check_one_of returns true if the current token's `Kind` is equal that of `expected_token`.
fn (mut p Parser) check_one_of(tokens []token.Kind) ? {
if p.tok.kind in tokens {
p.next() ?
} else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' expected one of $tokens but found "$p.tok.kind" in this (excerpt): "...${p.excerpt()}..."')
}
}
// is_at returns true if the token kind is equal to `expected_token`.
fn (mut p Parser) is_at(expected_token token.Kind) bool {
return p.tok.kind == expected_token
}
// expect will error if the token kind is not equal to `expected_token`.
fn (mut p Parser) expect(expected_token token.Kind) ? {
if p.tok.kind == expected_token {
return
} else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' expected token "$expected_token" but found "$p.tok.kind" in this text "...${p.excerpt()}..."')
}
}
2021-09-26 06:34:47 +02:00
// find_table returns a reference to a map if found in the *root* table given a "dotted" key (`a.b.c`).
2021-09-24 20:13:52 +02:00
// If some segments of the key does not exist in the root table find_table will
// allocate a new map for each segment. This behavior is needed because you can
// reference maps by multiple keys "dotted" (separated by "." periods) in TOML documents.
2021-09-26 06:34:47 +02:00
// See also `find_in_table`.
pub fn (mut p Parser) find_table() ?&map[string]ast.Value {
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'locating "$p.root_map_key" in map ${ptr_str(p.root_map)}')
mut t := &map[string]ast.Value{}
2021-09-24 20:13:52 +02:00
unsafe {
t = &p.root_map
}
if p.root_map_key == '' {
return t
}
return p.find_in_table(mut t, p.root_map_key)
}
2021-09-26 06:34:47 +02:00
// sub_table_key returns the logic parts of a dotted key (`a.b.c`) for
// use with the `find_sub_table` method.
2021-09-24 20:13:52 +02:00
pub fn (mut p Parser) sub_table_key(key string) (string, string) {
mut ks := key.split('.')
last := ks.last()
ks.delete_last()
return ks.join('.'), last
}
2021-09-26 06:34:47 +02:00
// find_sub_table returns a reference to a map if found in the *root* table given a "dotted" key (`a.b.c`).
// If some segments of the key does not exist in the input map find_sub_table will
2021-09-24 20:13:52 +02:00
// allocate a new map for the segment. This behavior is needed because you can
// reference maps by multiple keys "dotted" (separated by "." periods) in TOML documents.
2021-09-26 06:34:47 +02:00
// See also `find_in_table`.
pub fn (mut p Parser) find_sub_table(key string) ?&map[string]ast.Value {
2021-09-24 20:13:52 +02:00
mut ky := p.root_map_key + '.' + key
if p.root_map_key == '' {
ky = key
}
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'locating "$ky" in map ${ptr_str(p.root_map)}')
mut t := &map[string]ast.Value{}
2021-09-24 20:13:52 +02:00
unsafe {
t = &p.root_map
}
if ky == '' {
return t
}
return p.find_in_table(mut t, ky)
}
2021-09-26 06:34:47 +02:00
// find_in_table returns a reference to a map if found in `table` given a "dotted" key (`a.b.c`).
2021-09-24 20:13:52 +02:00
// If some segments of the key does not exist in the input map find_in_table will
// allocate a new map for the segment. This behavior is needed because you can
// reference maps by multiple keys "dotted" (separated by "." periods) in TOML documents.
pub fn (mut p Parser) find_in_table(mut table map[string]ast.Value, key string) ?&map[string]ast.Value {
2021-09-24 20:13:52 +02:00
// NOTE This code is the result of much trial and error.
// I'm still not quite sure *exactly* why it works. All I can leave here is a hope
// that this kind of minefield someday will be easier in V :)
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'locating "$key" in map ${ptr_str(table)}')
mut t := &map[string]ast.Value{}
2021-09-24 20:13:52 +02:00
unsafe {
t = &table
}
ks := key.split('.')
unsafe {
for k in ks {
if k in t.keys() {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'found key "$k" in $t.keys()')
if val := t[k] or {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' this should never happen. Key "$k" was checked before access')
}
{
if val is map[string]ast.Value {
2021-09-24 20:13:52 +02:00
// unsafe {
t = &(t[k] as map[string]ast.Value)
2021-09-24 20:13:52 +02:00
//}
} else {
return error(@MOD + '.' + @STRUCT + '.' + @FN + ' "$k" is not a map')
}
}
} else {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'no key "$k" found, allocating new map "$k" in map ${ptr_str(t)}"')
// unsafe {
t[k] = map[string]ast.Value{}
t = &(t[k] as map[string]ast.Value)
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'allocated new map ${ptr_str(t)}"')
//}
}
}
}
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'returning map ${ptr_str(t)}"')
return t
}
2021-09-26 06:34:47 +02:00
// sub_key parses next tokens as sub/nested keys. This is the also referred to as
// a "dotted" key (`a.b.c`). sub_key returns a string in dotted form.
2021-09-24 20:13:52 +02:00
pub fn (mut p Parser) sub_key() ?string {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing nested key...')
key := p.key() ?
mut text := key.str()
for p.peek_tok.kind == .period {
p.next() ? // .
p.check(.period) ?
next_key := p.key() ?
text += '.' + next_key.text
}
p.next() ?
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsed nested key `$text` now at "$p.tok.kind" "$p.tok.lit"')
return text
}
// root_table parses next tokens into the root map of `ast.Value`s.
2021-09-24 20:13:52 +02:00
// The V `map` type is corresponding to a "table" in TOML.
pub fn (mut p Parser) root_table() ? {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing root table...')
for p.tok.kind != .eof {
if !p.skip_next {
p.next() ?
} else {
p.skip_next = false
}
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing token "$p.tok.kind" "$p.tok.lit"')
match p.tok.kind {
.hash {
c := p.comment()
2021-10-26 15:58:05 +02:00
p.ast_root.comments << c
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'skipping comment "$c.text"')
}
//.whitespace, .tab, .nl {
// util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'skipping "$p.tok.kind "$p.tok.lit"')
//}
.bare, .quoted, .boolean, .number, .underscore { // NOTE .boolean allows for use of "true" and "false" as table keys
if p.peek_tok.kind == .assign
|| (p.tok.kind == .number && p.peek_tok.kind == .minus) {
key, val := p.key_value() ?
t := p.find_table() ?
unsafe {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'setting "$key.str()" = $val.to_json() in table ${ptr_str(t)}')
t[key.str()] = val
}
} else if p.peek_tok.kind == .period {
subkey := p.sub_key() ?
p.check(.assign) ?
val := p.value() ?
sub_table, key := p.sub_table_key(subkey)
t := p.find_sub_table(sub_table) ?
unsafe {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'setting "$key" = $val.to_json() in table ${ptr_str(t)}')
t[key] = val
}
} else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' dead end at "$p.tok.kind" "$p.tok.lit"')
}
}
.lsbr {
p.check(.lsbr) ? // '[' bracket
if p.tok.kind == .lsbr {
p.array_of_tables(mut &p.root_map) ?
p.skip_next = true // skip calling p.next() in coming iteration
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'leaving double bracket at "$p.tok.kind "$p.tok.lit". NEXT is "$p.peek_tok.kind "$p.peek_tok.lit"')
} else if p.peek_tok.kind == .period {
p.root_map_key = p.sub_key() ?
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'setting root map key to `$p.root_map_key` at "$p.tok.kind" "$p.tok.lit"')
p.expect(.rsbr) ?
} else {
key := p.key() ?
p.root_map_key = key.str()
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'setting root map key to `$p.root_map_key` at "$p.tok.kind" "$p.tok.lit"')
p.next() ?
p.expect(.rsbr) ?
}
}
.eof {
return
}
else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' could not parse "$p.tok.kind" "$p.tok.lit" in this (excerpt): "...${p.excerpt()}..."')
}
}
}
}
// excerpt returns a string of the characters surrounding `Parser.tok.pos`
fn (p Parser) excerpt() string {
return p.scanner.excerpt(p.tok.pos, 10)
}
// inline_table parses next tokens into a map of `ast.Value`s.
2021-09-24 20:13:52 +02:00
// The V map type is corresponding to a "table" in TOML.
pub fn (mut p Parser) inline_table(mut tbl map[string]ast.Value) ? {
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing inline table into ${ptr_str(tbl)}...')
mut previous_token_was_value := false
2021-09-24 20:13:52 +02:00
for p.tok.kind != .eof {
p.next() ?
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing token "$p.tok.kind"')
if previous_token_was_value {
if p.tok.kind != .rcbr {
p.expect(.comma) ?
2021-09-24 20:13:52 +02:00
}
previous_token_was_value = false
}
match p.tok.kind {
2021-09-24 20:13:52 +02:00
//.whitespace, .tab, .nl {
// util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'skipping "$p.tok.kind "$p.tok.lit"')
//}
.comma {
if p.peek_tok.kind == .rcbr {
p.next() ? // Forward to the peek_tok
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' unexpected "$p.tok.kind" "$p.tok.lit" at this (excerpt): "...${p.excerpt()}..."')
}
if p.peek_tok.kind == .comma {
p.next() ? // Forward to the peek_tok
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' unexpected "$p.tok.kind" "$p.tok.lit" at this (excerpt): "...${p.excerpt()}..."')
}
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'skipping comma table value seperator "$p.tok.lit"')
continue
}
.rcbr {
// '}' bracket
2021-09-24 20:13:52 +02:00
return
}
.bare, .quoted, .boolean, .number, .underscore {
if p.peek_tok.kind == .assign {
key, val := p.key_value() ?
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'inserting @5 "$key.str()" = $val.to_json() into ${ptr_str(tbl)}')
tbl[key.str()] = val
} else if p.peek_tok.kind == .period {
subkey := p.sub_key() ?
p.check(.assign) ?
val := p.value() ?
sub_table, key := p.sub_table_key(subkey)
mut t := p.find_in_table(mut tbl, sub_table) ?
unsafe {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'inserting @6 "$key" = $val.to_json() into ${ptr_str(t)}')
t[key] = val
}
} else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' dead end at "$p.tok.kind" "$p.tok.lit"')
}
previous_token_was_value = true
2021-09-24 20:13:52 +02:00
}
else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' unexpected "$p.tok.kind" "$p.tok.lit" at this (excerpt): "...${p.excerpt()}..."')
2021-09-24 20:13:52 +02:00
}
}
}
// Make sure the inline-table actually use the return at .rcbr match branch.
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' unexpected end of inline-table "$p.tok.kind" "$p.tok.lit" at this (excerpt): "...${p.excerpt()}..."')
2021-09-24 20:13:52 +02:00
}
// array_of_tables parses next tokens into an array of `ast.Value`s.
pub fn (mut p Parser) array_of_tables(mut table map[string]ast.Value) ? {
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing array of tables "$p.tok.kind" "$p.tok.lit"')
// NOTE this is starting to get ugly. TOML isn't simple at this point
p.check(.lsbr) ? // '[' bracket
// [[key.key]] horror
if p.peek_tok.kind == .period {
p.double_array_of_tables(mut table) ?
return
}
key := p.key() ?
p.next() ?
p.check(.rsbr) ?
p.check(.rsbr) ?
key_str := key.str()
unsafe {
if key_str in table.keys() {
if val := table[key_str] or {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' this should never happen. Key "$key_str" was checked before access')
}
{
if val is []ast.Value {
arr := &(table[key_str] as []ast.Value)
2021-09-24 20:13:52 +02:00
arr << p.double_bracket_array() ?
table[key_str] = arr
} else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' table[$key_str] is not an array. (excerpt): "...${p.excerpt()}..."')
}
}
} else {
table[key_str] = p.double_bracket_array() ?
}
}
p.last_aot = key_str
p.last_aot_index = 0
}
// double_array_of_tables parses next tokens into an array of tables of arrays of `ast.Value`s...
pub fn (mut p Parser) double_array_of_tables(mut table map[string]ast.Value) ? {
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing array of tables of arrays "$p.tok.kind" "$p.tok.lit"')
key := p.key() ?
mut key_str := key.str()
for p.peek_tok.kind == .period {
p.next() ? // .
p.check(.period) ?
next_key := p.key() ?
key_str += '.' + next_key.text
}
p.next() ?
p.check(.rsbr) ?
p.check(.rsbr) ?
ks := key_str.split('.')
if ks.len != 2 {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' nested array of tables does not support more than 2 levels. (excerpt): "...${p.excerpt()}..."')
}
first := ks[0]
last := ks[1]
unsafe {
// NOTE this is starting to get EVEN uglier. TOML is not at all simple at this point...
if p.last_aot != first {
table[first] = []ast.Value{}
2021-09-24 20:13:52 +02:00
p.last_aot = first
mut t_arr := &(table[p.last_aot] as []ast.Value)
t_arr << map[string]ast.Value{}
2021-09-24 20:13:52 +02:00
p.last_aot_index = 0
}
mut t_arr := &(table[p.last_aot] as []ast.Value)
2021-09-24 20:13:52 +02:00
mut t_map := t_arr[p.last_aot_index]
mut t := &(t_map as map[string]ast.Value)
2021-09-24 20:13:52 +02:00
if last in t.keys() {
if val := t[last] or {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' this should never happen. Key "$last" was checked before access')
}
{
if val is []ast.Value {
arr := &(val as []ast.Value)
2021-09-24 20:13:52 +02:00
arr << p.double_bracket_array() ?
t[last] = arr
} else {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' t[$last] is not an array. (excerpt): "...${p.excerpt()}..."')
}
}
} else {
t[last] = p.double_bracket_array() ?
}
}
}
// array parses next tokens into an array of `ast.Value`s.
pub fn (mut p Parser) double_bracket_array() ?[]ast.Value {
mut arr := []ast.Value{}
2021-09-24 20:13:52 +02:00
for p.tok.kind in [.bare, .quoted, .boolean, .number] && p.peek_tok.kind == .assign {
mut tbl := map[string]ast.Value{}
2021-09-24 20:13:52 +02:00
key, val := p.key_value() ?
tbl[key.str()] = val
arr << tbl
p.next() ?
}
return arr
}
// array parses next tokens into an array of `ast.Value`s.
pub fn (mut p Parser) array() ?[]ast.Value {
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing array...')
mut arr := []ast.Value{}
2021-09-24 20:13:52 +02:00
p.expect(.lsbr) ? // '[' bracket
2021-10-27 19:26:33 +02:00
mut previous_token_was_value := false
2021-09-24 20:13:52 +02:00
for p.tok.kind != .eof {
p.next() ?
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing token "$p.tok.kind" "$p.tok.lit"')
2021-10-27 19:26:33 +02:00
if previous_token_was_value {
if p.tok.kind != .rsbr && p.tok.kind != .hash {
p.expect(.comma) ?
}
previous_token_was_value = false
}
2021-09-24 20:13:52 +02:00
match p.tok.kind {
.boolean {
arr << ast.Value(p.boolean() ?)
2021-10-27 19:26:33 +02:00
previous_token_was_value = true
2021-09-24 20:13:52 +02:00
}
.comma {
2021-10-27 19:26:33 +02:00
// Trailing commas before array close is allowed
// so we skip `if p.peek_tok.kind == .rsbr { ... }`
if p.peek_tok.kind == .comma {
p.next() ? // Forward to the peek_tok
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' unexpected "$p.tok.kind" "$p.tok.lit" at this (excerpt): "...${p.excerpt()}..."')
}
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'skipping comma table value seperator "$p.tok.lit"')
2021-09-24 20:13:52 +02:00
continue
}
.eof {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' could not parse array. Reached EOF "$p.tok.kind" "$p.tok.lit" ("$p.tok.lit") in this (excerpt): "...${p.excerpt()}..."')
2021-09-24 20:13:52 +02:00
}
.hash {
c := p.comment()
2021-10-27 19:26:33 +02:00
p.ast_root.comments << c
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'skipping comment "$c.text"')
}
.lcbr {
mut t := map[string]ast.Value{}
2021-09-24 20:13:52 +02:00
p.inline_table(mut t) ?
2021-10-27 19:26:33 +02:00
arr << ast.Value(t)
previous_token_was_value = true
2021-09-24 20:13:52 +02:00
}
.number {
val := p.number_or_date() ?
arr << val
2021-10-27 19:26:33 +02:00
previous_token_was_value = true
2021-09-24 20:13:52 +02:00
}
.quoted {
arr << ast.Value(p.quoted())
2021-10-27 19:26:33 +02:00
previous_token_was_value = true
2021-09-24 20:13:52 +02:00
}
.lsbr {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing array in array "$p.tok.kind" "$p.tok.lit"')
arr << ast.Value(p.array() ?)
2021-10-27 19:26:33 +02:00
previous_token_was_value = true
2021-09-24 20:13:52 +02:00
}
.rsbr {
break
}
else {
error(@MOD + '.' + @STRUCT + '.' + @FN +
' could not parse "$p.tok.kind" "$p.tok.lit" ("$p.tok.lit") in this (excerpt): "...${p.excerpt()}..."')
}
}
}
p.expect(.rsbr) ? // ']' bracket
$if debug {
flat := arr.str().replace('\n', r'\n')
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsed array: $flat . Currently @ token "$p.tok.kind"')
}
return arr
}
// comment returns an `ast.Comment` type.
pub fn (mut p Parser) comment() ast.Comment {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsed hash comment "#$p.tok.lit"')
return ast.Comment{
text: p.tok.lit
pos: p.tok.position()
}
}
// key parse and returns an `ast.Key` type.
// Keys are the token(s) appearing before an assignment operator (=).
pub fn (mut p Parser) key() ?ast.Key {
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing key from "$p.tok.lit" ...')
mut key := ast.Key(ast.Null{})
if p.tok.kind == .number {
if p.peek_tok.kind == .minus {
mut lits := p.tok.lit
pos := p.tok.position()
for p.peek_tok.kind != .assign {
p.next() ?
lits += p.tok.lit
}
return ast.Key(ast.Bare{
text: lits
pos: pos
})
}
// number := p.number() as ast.Number
key = ast.Key(p.number())
} else {
key = match p.tok.kind {
.bare, .underscore {
ast.Key(p.bare())
}
.boolean {
ast.Key(p.boolean() ?)
}
.quoted {
ast.Key(p.quoted())
}
else {
error(@MOD + '.' + @STRUCT + '.' + @FN +
' key expected .bare, .number, .quoted or .boolean but got "$p.tok.kind"')
ast.Key(ast.Bare{}) // TODO workaround bug
}
}
}
// NOTE kept for eased debugging
// util.printdbg(@MOD +'.' + @STRUCT + '.' + @FN, 'parsed key "$p.tok.lit"')
// panic(@MOD + '.' + @STRUCT + '.' + @FN + ' could not parse ${p.tok.kind} ("${p.tok.lit}") token \n$p.tok')
// return ast.Key(ast.Bare{})
return key
}
// key_value parse and returns a pair `ast.Key` and `ast.Value` type.
2021-09-24 20:13:52 +02:00
// see also `key()` and `value()`
pub fn (mut p Parser) key_value() ?(ast.Key, ast.Value) {
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing key value pair...')
key := p.key() ?
p.next() ?
p.check(.assign) ? // Assignment operator
value := p.value() ?
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsed key value pair. "$key" = $value.to_json()')
return key, value
}
// value parse and returns an `ast.Value` type.
2021-09-24 20:13:52 +02:00
// values are the token(s) appearing after an assignment operator (=).
pub fn (mut p Parser) value() ?ast.Value {
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing value...')
// println('parsed comment "${p.tok.lit}"')
mut value := ast.Value(ast.Null{})
2021-09-24 20:13:52 +02:00
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsing token "$p.tok.kind" "$p.tok.lit"')
// mut value := ast.Value{}
2021-09-24 20:13:52 +02:00
if p.tok.kind == .number {
number_or_date := p.number_or_date() ?
value = number_or_date
} else {
value = match p.tok.kind {
.quoted {
ast.Value(p.quoted())
2021-09-24 20:13:52 +02:00
}
.boolean {
ast.Value(p.boolean() ?)
2021-09-24 20:13:52 +02:00
}
.lsbr {
ast.Value(p.array() ?)
2021-09-24 20:13:52 +02:00
}
.lcbr {
mut t := map[string]ast.Value{}
2021-09-24 20:13:52 +02:00
p.inline_table(mut t) ?
// table[key_str] = ast.Value(t)
ast.Value(t)
2021-09-24 20:13:52 +02:00
}
else {
ast.Value(ast.Null{})
2021-09-24 20:13:52 +02:00
}
}
if value is ast.Null {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' value expected .boolean, .quoted, .lsbr, .lcbr or .number got "$p.tok.kind" "$p.tok.lit"')
}
2021-09-24 20:13:52 +02:00
}
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsed "$p.tok.kind" as value $value.to_json()')
2021-09-24 20:13:52 +02:00
return value
}
// number_or_date parse and returns an `ast.Value` type as
2021-09-24 20:13:52 +02:00
// one of [`ast.Date`, `ast.Time`, `ast.DateTime`, `ast.Number`]
pub fn (mut p Parser) number_or_date() ?ast.Value {
2021-09-24 20:13:52 +02:00
// Handle Date/Time
if p.peek_tok.kind == .minus || p.peek_tok.kind == .colon {
date_time_type := p.date_time() ?
match date_time_type {
ast.Date {
return ast.Value(date_time_type as ast.Date)
2021-09-24 20:13:52 +02:00
}
ast.Time {
return ast.Value(date_time_type as ast.Time)
2021-09-24 20:13:52 +02:00
}
ast.DateTime {
return ast.Value(date_time_type as ast.DateTime)
2021-09-24 20:13:52 +02:00
}
}
}
return ast.Value(p.number())
2021-09-24 20:13:52 +02:00
}
// bare parse and returns an `ast.Bare` type.
pub fn (mut p Parser) bare() ast.Bare {
return ast.Bare{
text: p.tok.lit
pos: p.tok.position()
}
}
// quoted parse and returns an `ast.Quoted` type.
pub fn (mut p Parser) quoted() ast.Quoted {
// To get more info about the quote type and enable better checking,
// the scanner is returning the literal *with* single- or double-quotes.
mut quote := p.tok.lit[0]
is_multiline := p.tok.lit.len >= 6 && p.tok.lit[1] == quote && p.tok.lit[2] == quote
mut lit := p.tok.lit[1..p.tok.lit.len - 1]
if is_multiline {
lit = p.tok.lit[3..p.tok.lit.len - 3]
}
2021-09-24 20:13:52 +02:00
return ast.Quoted{
text: lit
2021-09-24 20:13:52 +02:00
pos: p.tok.position()
quote: quote
is_multiline: is_multiline
2021-09-24 20:13:52 +02:00
}
}
// boolean parse and returns an `ast.Bool` type.
pub fn (mut p Parser) boolean() ?ast.Bool {
if p.tok.lit !in ['true', 'false'] {
return error(@MOD + '.' + @STRUCT + '.' + @FN +
' expected literal to be either `true` or `false` got "$p.tok.kind"')
}
return ast.Bool{
text: p.tok.lit
pos: p.tok.position()
}
}
// number parse and returns an `ast.Number` type.
pub fn (mut p Parser) number() ast.Number {
return ast.Number{
text: p.tok.lit
pos: p.tok.position()
}
}
// date_time parses dates and time in RFC 3339 format.
// https://datatracker.ietf.org/doc/html/rfc3339
pub fn (mut p Parser) date_time() ?ast.DateTimeType {
// Date and/or Time
mut lit := ''
pos := p.tok.position()
mut date := ast.Date{}
mut time := ast.Time{}
if p.peek_tok.kind == .minus {
date = p.date() ?
lit += date.text
// Look for any THH:MM:SS or <space>HH:MM:SS
if (p.peek_tok.kind == .bare && (p.peek_tok.lit.starts_with('T')
|| p.peek_tok.lit.starts_with('t'))) || p.peek_tok.kind == .whitespace {
p.next() ? // Advance to token with Txx or whitespace special case
if p.tok.lit.starts_with('T') || p.tok.lit.starts_with('t') {
lit += p.tok.lit[0].ascii_str() //'T' or 't'
} else {
lit += p.tok.lit
p.next() ?
}
time = p.time() ?
lit += time.text
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsed date-time: "$lit"')
return ast.DateTime{
text: lit
pos: pos
date: date
time: time
}
}
} else if p.peek_tok.kind == .colon {
time = p.time() ?
return time
}
return ast.Date{
text: lit
pos: pos
}
}
// date parse and returns an `ast.Date` type.
pub fn (mut p Parser) date() ?ast.Date {
// Date
mut lit := p.tok.lit
pos := p.tok.position()
p.check(.number) ?
lit += p.tok.lit
p.check(.minus) ?
lit += p.tok.lit
p.check(.number) ?
lit += p.tok.lit
p.check(.minus) ?
lit += p.tok.lit
p.expect(.number) ?
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsed date: "$lit"')
return ast.Date{
text: lit
pos: pos
}
}
// time parse and returns an `ast.Time` type.
pub fn (mut p Parser) time() ?ast.Time {
// Time
mut lit := p.tok.lit
pos := p.tok.position()
if p.is_at(.bare) && (lit.starts_with('T') || lit.starts_with('t')) {
if p.tok.lit.starts_with('T') {
lit = lit.all_after('T')
} else if p.tok.lit.starts_with('t') {
lit = lit.all_after('t')
}
p.next() ?
} else {
p.check(.number) ?
}
lit += p.tok.lit
p.check(.colon) ?
lit += p.tok.lit
p.check(.number) ?
lit += p.tok.lit
// TODO does TOML even have optional seconds?
// if p.peek_tok.kind == .colon {
p.check(.colon) ?
lit += p.tok.lit
p.expect(.number) ?
//}
// Optional milliseconds
if p.peek_tok.kind == .period {
p.next() ?
lit += p.tok.lit // lit += '.'
p.check(.period) ?
lit += p.tok.lit
p.expect(.number) ?
}
// Parse offset
if p.peek_tok.kind == .minus || p.peek_tok.kind == .plus {
p.next() ?
lit += p.tok.lit // lit += '-'
p.check_one_of([.minus, .plus]) ?
lit += p.tok.lit
p.check(.number) ?
lit += p.tok.lit
p.check(.colon) ?
lit += p.tok.lit
p.expect(.number) ?
} else if p.peek_tok.kind == .bare && (p.peek_tok.lit == 'Z' || p.peek_tok.lit == 'z') {
p.next() ?
lit += p.tok.lit
p.expect(.bare) ?
}
util.printdbg(@MOD + '.' + @STRUCT + '.' + @FN, 'parsed time: "$lit"')
return ast.Time{
text: lit
pos: pos
}
}
// eof returns an `ast.EOF` type.
pub fn (mut p Parser) eof() ast.EOF {
return ast.EOF{
pos: p.tok.position()
}
}