322 lines
9.9 KiB
V
322 lines
9.9 KiB
V
import os
|
|
import toml
|
|
import toml.ast
|
|
import toml.scanner
|
|
import x.json2
|
|
import strconv
|
|
|
|
// Instructions for developers:
|
|
// The actual tests and data can be obtained by doing:
|
|
// `cd vlib/toml/tests/testdata`
|
|
// `git clone --depth 1 https://github.com/BurntSushi/toml-test.git burntsushi/toml-test`
|
|
// See also the CI toml tests
|
|
// TODO Goal: make value retrieval of all of https://github.com/BurntSushi/toml-test/test/ pass
|
|
const (
|
|
// Kept for easier handling of future updates to the tests
|
|
valid_exceptions = []string{}
|
|
invalid_exceptions = []string{}
|
|
|
|
valid_value_exceptions = [
|
|
// String
|
|
'string/escapes.toml',
|
|
'string/escape-tricky.toml',
|
|
'string/multiline.toml',
|
|
// Integer
|
|
'integer/long.toml',
|
|
// Float
|
|
'float/inf-and-nan.toml',
|
|
// Comment
|
|
'comment/tricky.toml',
|
|
// Table
|
|
'table/array-implicit.toml',
|
|
'table/names.toml',
|
|
// Date-time
|
|
'datetime/milliseconds.toml',
|
|
// Inline-table
|
|
'inline-table/multiline.toml',
|
|
// Key
|
|
'key/numeric-dotted.toml',
|
|
'key/alphanum.toml',
|
|
'key/escapes.toml',
|
|
]
|
|
|
|
jq = os.find_abs_path_of_executable('jq') or { '' }
|
|
compare_work_dir_root = os.join_path(os.temp_dir(), 'v', 'toml', 'burntsushi')
|
|
// From: https://stackoverflow.com/a/38266731/1904615
|
|
jq_normalize = r'# Apply f to composite entities recursively using keys[], and to atoms
|
|
def sorted_walk(f):
|
|
. as $in
|
|
| if type == "object" then
|
|
reduce keys[] as $key
|
|
( {}; . + { ($key): ($in[$key] | sorted_walk(f)) } ) | f
|
|
elif type == "array" then map( sorted_walk(f) ) | f
|
|
else f
|
|
end;
|
|
|
|
def normalize: sorted_walk(if type == "array" then sort else . end);
|
|
|
|
normalize'
|
|
)
|
|
|
|
fn run(args []string) ?string {
|
|
res := os.execute(args.join(' '))
|
|
if res.exit_code != 0 {
|
|
return error('${args[0]} failed with return code ${res.exit_code}.\n$res.output')
|
|
}
|
|
return res.output
|
|
}
|
|
|
|
// test_burnt_sushi_tomltest run though 'testdata/burntsushi/toml-test/*' if found.
|
|
fn test_burnt_sushi_tomltest() {
|
|
this_file := @FILE
|
|
test_root := os.join_path(os.dir(this_file), 'testdata', 'burntsushi', 'toml-test',
|
|
'tests')
|
|
if os.is_dir(test_root) {
|
|
valid_test_files := os.walk_ext(os.join_path(test_root, 'valid'), '.toml')
|
|
println('Testing $valid_test_files.len valid TOML files...')
|
|
mut valid := 0
|
|
mut e := 0
|
|
for i, valid_test_file in valid_test_files {
|
|
mut relative := valid_test_file.all_after(os.join_path('toml-test', 'tests',
|
|
'valid')).trim_left(os.path_separator)
|
|
$if windows {
|
|
relative = relative.replace('/', '\\')
|
|
}
|
|
if relative !in valid_exceptions {
|
|
println('OK [${i + 1}/$valid_test_files.len] "$valid_test_file"...')
|
|
toml_doc := toml.parse_file(valid_test_file) or { panic(err) }
|
|
valid++
|
|
} else {
|
|
e++
|
|
println('SKIP [${i + 1}/$valid_test_files.len] "$valid_test_file" EXCEPTION [$e/$valid_exceptions.len]...')
|
|
}
|
|
}
|
|
println('$valid/$valid_test_files.len TOML files was parsed correctly')
|
|
if valid_exceptions.len > 0 {
|
|
println('TODO Skipped parsing of $valid_exceptions.len valid TOML files...')
|
|
}
|
|
|
|
// If the command-line tool `jq` is installed, value tests can be run as well.
|
|
if jq != '' {
|
|
println('Testing value output of $valid_test_files.len valid TOML files using "$jq"...')
|
|
|
|
if os.exists(compare_work_dir_root) {
|
|
os.rmdir_all(compare_work_dir_root) or { panic(err) }
|
|
}
|
|
os.mkdir_all(compare_work_dir_root) or { panic(err) }
|
|
|
|
jq_normalize_path := os.join_path(compare_work_dir_root, 'normalize.jq')
|
|
os.write_file(jq_normalize_path, jq_normalize) or { panic(err) }
|
|
|
|
valid = 0
|
|
e = 0
|
|
for i, valid_test_file in valid_test_files {
|
|
mut relative := valid_test_file.all_after(os.join_path('toml-test', 'tests',
|
|
'valid')).trim_left(os.path_separator)
|
|
$if windows {
|
|
relative = relative.replace('/', '\\')
|
|
}
|
|
// Skip the file if we know it can't be parsed or we know that the value retrieval needs work.
|
|
if relative !in valid_exceptions && relative !in valid_value_exceptions {
|
|
println('OK [${i + 1}/$valid_test_files.len] "$valid_test_file"...')
|
|
toml_doc := toml.parse_file(valid_test_file) or { panic(err) }
|
|
|
|
v_toml_json_path := os.join_path(compare_work_dir_root,
|
|
os.file_name(valid_test_file).all_before_last('.') + '.v.json')
|
|
bs_toml_json_path := os.join_path(compare_work_dir_root,
|
|
os.file_name(valid_test_file).all_before_last('.') + '.json')
|
|
|
|
os.write_file(v_toml_json_path, to_burntsushi(toml_doc.ast.table)) or {
|
|
panic(err)
|
|
}
|
|
|
|
bs_json := os.read_file(valid_test_file.all_before_last('.') + '.json') or {
|
|
panic(err)
|
|
}
|
|
os.write_file(bs_toml_json_path, bs_json) or { panic(err) }
|
|
|
|
v_normalized_json := run([jq, '-S', '-f "$jq_normalize_path"', v_toml_json_path]) or {
|
|
contents := os.read_file(v_toml_json_path) or { panic(err) }
|
|
panic(err.msg + '\n$contents')
|
|
}
|
|
bs_normalized_json := run([jq, '-S', '-f "$jq_normalize_path"', bs_toml_json_path]) or {
|
|
contents := os.read_file(v_toml_json_path) or { panic(err) }
|
|
panic(err.msg + '\n$contents')
|
|
}
|
|
|
|
assert bs_normalized_json == v_normalized_json
|
|
|
|
valid++
|
|
} else {
|
|
e++
|
|
println('SKIP [${i + 1}/$valid_test_files.len] "$valid_test_file" EXCEPTION [$e/$valid_value_exceptions.len]...')
|
|
}
|
|
}
|
|
println('$valid/$valid_test_files.len TOML files was parsed correctly and value checked')
|
|
if valid_value_exceptions.len > 0 {
|
|
println('TODO Skipped value checks of $valid_value_exceptions.len valid TOML files...')
|
|
}
|
|
}
|
|
|
|
// TODO test cases where the parser should fail
|
|
invalid_test_files := os.walk_ext(os.join_path(test_root, 'invalid'), '.toml')
|
|
println('Testing $invalid_test_files.len invalid TOML files...')
|
|
mut invalid := 0
|
|
e = 0
|
|
for i, invalid_test_file in invalid_test_files {
|
|
mut relative := invalid_test_file.all_after(os.join_path('toml-test', 'tests',
|
|
'invalid')).trim_left(os.path_separator)
|
|
$if windows {
|
|
relative = relative.replace('/', '\\')
|
|
}
|
|
if relative !in invalid_exceptions {
|
|
println('OK [${i + 1}/$invalid_test_files.len] "$invalid_test_file"...')
|
|
if toml_doc := toml.parse_file(invalid_test_file) {
|
|
content_that_should_have_failed := os.read_file(invalid_test_file) or {
|
|
panic(err)
|
|
}
|
|
println(' This TOML should have failed:\n${'-'.repeat(40)}\n$content_that_should_have_failed\n${'-'.repeat(40)}')
|
|
assert false
|
|
} else {
|
|
println(' $err.msg')
|
|
assert true
|
|
}
|
|
invalid++
|
|
} else {
|
|
e++
|
|
println('SKIP [${i + 1}/$invalid_test_files.len] "$invalid_test_file" EXCEPTION [$e/$invalid_exceptions.len]...')
|
|
}
|
|
}
|
|
println('$invalid/$invalid_test_files.len TOML files was parsed correctly')
|
|
if invalid_exceptions.len > 0 {
|
|
println('TODO Skipped parsing of $invalid_exceptions.len invalid TOML files...')
|
|
}
|
|
} else {
|
|
println('No test data directory found in "$test_root"')
|
|
assert true
|
|
}
|
|
}
|
|
|
|
// to_burntsushi returns a BurntSushi compatible json string converted from the `value` ast.Value.
|
|
fn to_burntsushi(value ast.Value) string {
|
|
match value {
|
|
ast.Quoted {
|
|
mut json_text := ''
|
|
if value.quote == `"` {
|
|
json_text = toml_to_json_escapes(value) or { '<error>' }
|
|
} else {
|
|
json_text = json2.Any(value.text).json_str()
|
|
}
|
|
|
|
return '{ "type": "string", "value": "$json_text" }'
|
|
}
|
|
ast.DateTime {
|
|
// Normalization for json
|
|
json_text := json2.Any(value.text).json_str().to_upper().replace(' ', 'T')
|
|
typ := if json_text.ends_with('Z') || json_text.all_after('T').contains('-')
|
|
|| json_text.all_after('T').contains('+') {
|
|
'datetime'
|
|
} else {
|
|
'datetime-local'
|
|
}
|
|
return '{ "type": "$typ", "value": "$json_text" }'
|
|
}
|
|
ast.Date {
|
|
json_text := json2.Any(value.text).json_str()
|
|
return '{ "type": "date-local", "value": "$json_text" }'
|
|
}
|
|
ast.Time {
|
|
json_text := json2.Any(value.text).json_str()
|
|
return '{ "type": "time-local", "value": "$json_text" }'
|
|
}
|
|
ast.Bool {
|
|
json_text := json2.Any(value.text.bool()).json_str()
|
|
return '{ "type": "bool", "value": "$json_text" }'
|
|
}
|
|
ast.Null {
|
|
json_text := json2.Any(value.text).json_str()
|
|
return '{ "type": "null", "value": "$json_text" }'
|
|
}
|
|
ast.Number {
|
|
if value.text.contains('inf') || value.text.contains('nan') {
|
|
return '{ "type": "float", "value": "$value.text" }'
|
|
}
|
|
if !value.text.starts_with('0x')
|
|
&& (value.text.contains('.') || value.text.to_lower().contains('e')) {
|
|
mut val := '$value.f64()'.replace('.e+', '.0e') // json notation
|
|
if !val.contains('.') && val != '0' { // json notation
|
|
val += '.0'
|
|
}
|
|
return '{ "type": "float", "value": "$val" }'
|
|
}
|
|
return '{ "type": "integer", "value": "$value.i64()" }'
|
|
}
|
|
map[string]ast.Value {
|
|
mut str := '{ '
|
|
for key, val in value {
|
|
json_key := json2.Any(key).json_str()
|
|
str += ' "$json_key": ${to_burntsushi(val)},'
|
|
}
|
|
str = str.trim_right(',')
|
|
str += ' }'
|
|
return str
|
|
}
|
|
[]ast.Value {
|
|
mut str := '[ '
|
|
for val in value {
|
|
str += ' ${to_burntsushi(val)},'
|
|
}
|
|
str = str.trim_right(',')
|
|
str += ' ]\n'
|
|
return str
|
|
}
|
|
}
|
|
return '<error>'
|
|
}
|
|
|
|
// toml_to_json_escapes is a utility function for normalizing
|
|
// TOML basic string to JSON string
|
|
fn toml_to_json_escapes(q ast.Quoted) ?string {
|
|
mut s := scanner.new_simple(q.text) ?
|
|
mut r := ''
|
|
for {
|
|
ch := s.next()
|
|
if ch == scanner.end_of_text {
|
|
break
|
|
}
|
|
ch_byte := byte(ch)
|
|
|
|
if ch == `"` {
|
|
if byte(s.peek(-1)) != `\\` {
|
|
r += '\\'
|
|
}
|
|
}
|
|
|
|
if ch == `\\` {
|
|
next_ch := byte(s.at())
|
|
|
|
escape := ch_byte.ascii_str() + next_ch.ascii_str()
|
|
if escape.to_lower() == '\\u' {
|
|
mut b := s.next()
|
|
mut unicode_point := ''
|
|
for {
|
|
b = s.next()
|
|
if b != ` ` && b != scanner.end_of_text {
|
|
unicode_point += byte(b).ascii_str()
|
|
} else {
|
|
break
|
|
}
|
|
}
|
|
if unicode_point.len < 8 {
|
|
unicode_point = '0'.repeat(8 - unicode_point.len) + unicode_point
|
|
}
|
|
rn := rune(strconv.parse_int(unicode_point, 16, 0) ?)
|
|
r += '$rn'
|
|
continue
|
|
}
|
|
}
|
|
r += ch_byte.ascii_str()
|
|
}
|
|
return r
|
|
}
|