mirror of
https://github.com/cachix/install-nix-action.git
synced 2024-11-22 16:40:51 +00:00
303 lines
7.3 KiB
JavaScript
303 lines
7.3 KiB
JavaScript
|
"use strict";
|
||
|
const whatwgEncoding = require("whatwg-encoding");
|
||
|
|
||
|
// https://html.spec.whatwg.org/#encoding-sniffing-algorithm
|
||
|
module.exports = function sniffHTMLEncoding(buffer, options) {
|
||
|
let encoding = whatwgEncoding.getBOMEncoding(buffer); // see https://github.com/whatwg/html/issues/1910
|
||
|
|
||
|
if (options === undefined) {
|
||
|
options = {};
|
||
|
}
|
||
|
|
||
|
if (encoding === null && options.transportLayerEncodingLabel !== undefined) {
|
||
|
encoding = whatwgEncoding.labelToName(options.transportLayerEncodingLabel);
|
||
|
}
|
||
|
|
||
|
if (encoding === null) {
|
||
|
encoding = prescanMetaCharset(buffer);
|
||
|
}
|
||
|
|
||
|
if (encoding === null && options.defaultEncoding !== undefined) {
|
||
|
encoding = options.defaultEncoding;
|
||
|
}
|
||
|
|
||
|
if (encoding === null) {
|
||
|
encoding = "windows-1252";
|
||
|
}
|
||
|
|
||
|
return encoding;
|
||
|
};
|
||
|
|
||
|
// https://html.spec.whatwg.org/multipage/syntax.html#prescan-a-byte-stream-to-determine-its-encoding
|
||
|
function prescanMetaCharset(buffer) {
|
||
|
const l = Math.min(buffer.length, 1024);
|
||
|
for (let i = 0; i < l; i++) {
|
||
|
let c = buffer[i];
|
||
|
if (c === 0x3C) {
|
||
|
// "<"
|
||
|
let c1 = buffer[i + 1];
|
||
|
let c2 = buffer[i + 2];
|
||
|
const c3 = buffer[i + 3];
|
||
|
const c4 = buffer[i + 4];
|
||
|
const c5 = buffer[i + 5];
|
||
|
// !-- (comment start)
|
||
|
if (c1 === 0x21 && c2 === 0x2D && c3 === 0x2D) {
|
||
|
i += 4;
|
||
|
for (; i < l; i++) {
|
||
|
c = buffer[i];
|
||
|
c1 = buffer[i + 1];
|
||
|
c2 = buffer[i + 2];
|
||
|
// --> (comment end)
|
||
|
if (c === 0x2D && c1 === 0x2D && c2 === 0x3E) {
|
||
|
i += 2;
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
} else if ((c1 === 0x4D || c1 === 0x6D) &&
|
||
|
(c2 === 0x45 || c2 === 0x65) &&
|
||
|
(c3 === 0x54 || c3 === 0x74) &&
|
||
|
(c4 === 0x41 || c4 === 0x61) &&
|
||
|
(isSpaceCharacter(c5) || c5 === 0x2F)) {
|
||
|
// "meta" + space or /
|
||
|
i += 6;
|
||
|
let gotPragma = false;
|
||
|
let needPragma = null;
|
||
|
let charset = null;
|
||
|
|
||
|
let attrRes;
|
||
|
do {
|
||
|
attrRes = getAttribute(buffer, i, l);
|
||
|
if (attrRes.attr) {
|
||
|
if (attrRes.attr.name === "http-equiv") {
|
||
|
gotPragma = attrRes.attr.value === "content-type";
|
||
|
} else if (attrRes.attr.name === "content" && !charset) {
|
||
|
charset = extractCharacterEncodingFromMeta(attrRes.attr.value);
|
||
|
if (charset !== null) {
|
||
|
needPragma = true;
|
||
|
}
|
||
|
} else if (attrRes.attr.name === "charset") {
|
||
|
charset = whatwgEncoding.labelToName(attrRes.attr.value);
|
||
|
needPragma = false;
|
||
|
}
|
||
|
}
|
||
|
i = attrRes.i;
|
||
|
} while (attrRes.attr);
|
||
|
|
||
|
if (needPragma === null) {
|
||
|
continue;
|
||
|
}
|
||
|
if (needPragma === true && gotPragma === false) {
|
||
|
continue;
|
||
|
}
|
||
|
if (charset === null) {
|
||
|
continue;
|
||
|
}
|
||
|
|
||
|
if (charset === "UTF-16LE" || charset === "UTF-16BE") {
|
||
|
charset = "UTF-8";
|
||
|
}
|
||
|
if (charset === "x-user-defined") {
|
||
|
charset = "windows-1252";
|
||
|
}
|
||
|
|
||
|
return charset;
|
||
|
} else if ((c1 >= 0x41 && c1 <= 0x5A) || (c1 >= 0x61 && c1 <= 0x7A)) {
|
||
|
// a-z or A-Z
|
||
|
for (i += 2; i < l; i++) {
|
||
|
c = buffer[i];
|
||
|
// space or >
|
||
|
if (isSpaceCharacter(c) || c === 0x3E) {
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
let attrRes;
|
||
|
do {
|
||
|
attrRes = getAttribute(buffer, i, l);
|
||
|
i = attrRes.i;
|
||
|
} while (attrRes.attr);
|
||
|
} else if (c1 === 0x21 || c1 === 0x2F || c1 === 0x3F) {
|
||
|
// ! or / or ?
|
||
|
for (i += 2; i < l; i++) {
|
||
|
c = buffer[i];
|
||
|
// >
|
||
|
if (c === 0x3E) {
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
return null;
|
||
|
}
|
||
|
|
||
|
// https://html.spec.whatwg.org/multipage/syntax.html#concept-get-attributes-when-sniffing
|
||
|
function getAttribute(buffer, i, l) {
|
||
|
for (; i < l; i++) {
|
||
|
let c = buffer[i];
|
||
|
// space or /
|
||
|
if (isSpaceCharacter(c) || c === 0x2F) {
|
||
|
continue;
|
||
|
}
|
||
|
// ">"
|
||
|
if (c === 0x3E) {
|
||
|
i++;
|
||
|
break;
|
||
|
}
|
||
|
let name = "";
|
||
|
let value = "";
|
||
|
nameLoop:for (; i < l; i++) {
|
||
|
c = buffer[i];
|
||
|
// "="
|
||
|
if (c === 0x3D && name !== "") {
|
||
|
i++;
|
||
|
break;
|
||
|
}
|
||
|
// space
|
||
|
if (isSpaceCharacter(c)) {
|
||
|
for (i++; i < l; i++) {
|
||
|
c = buffer[i];
|
||
|
// space
|
||
|
if (isSpaceCharacter(c)) {
|
||
|
continue;
|
||
|
}
|
||
|
// not "="
|
||
|
if (c !== 0x3D) {
|
||
|
return { attr: { name, value }, i };
|
||
|
}
|
||
|
|
||
|
i++;
|
||
|
break nameLoop;
|
||
|
}
|
||
|
break;
|
||
|
}
|
||
|
// / or >
|
||
|
if (c === 0x2F || c === 0x3E) {
|
||
|
return { attr: { name, value }, i };
|
||
|
}
|
||
|
// A-Z
|
||
|
if (c >= 0x41 && c <= 0x5A) {
|
||
|
name += String.fromCharCode(c + 0x20); // lowercase
|
||
|
} else {
|
||
|
name += String.fromCharCode(c);
|
||
|
}
|
||
|
}
|
||
|
c = buffer[i];
|
||
|
// space
|
||
|
if (isSpaceCharacter(c)) {
|
||
|
for (i++; i < l; i++) {
|
||
|
c = buffer[i];
|
||
|
// space
|
||
|
if (isSpaceCharacter(c)) {
|
||
|
continue;
|
||
|
} else {
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
// " or '
|
||
|
if (c === 0x22 || c === 0x27) {
|
||
|
const quote = c;
|
||
|
for (i++; i < l; i++) {
|
||
|
c = buffer[i];
|
||
|
|
||
|
if (c === quote) {
|
||
|
i++;
|
||
|
return { attr: { name, value }, i };
|
||
|
}
|
||
|
|
||
|
// A-Z
|
||
|
if (c >= 0x41 && c <= 0x5A) {
|
||
|
value += String.fromCharCode(c + 0x20); // lowercase
|
||
|
} else {
|
||
|
value += String.fromCharCode(c);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
// >
|
||
|
if (c === 0x3E) {
|
||
|
return { attr: { name, value }, i };
|
||
|
}
|
||
|
|
||
|
// A-Z
|
||
|
if (c >= 0x41 && c <= 0x5A) {
|
||
|
value += String.fromCharCode(c + 0x20); // lowercase
|
||
|
} else {
|
||
|
value += String.fromCharCode(c);
|
||
|
}
|
||
|
|
||
|
for (i++; i < l; i++) {
|
||
|
c = buffer[i];
|
||
|
|
||
|
// space or >
|
||
|
if (isSpaceCharacter(c) || c === 0x3E) {
|
||
|
return { attr: { name, value }, i };
|
||
|
}
|
||
|
|
||
|
// A-Z
|
||
|
if (c >= 0x41 && c <= 0x5A) {
|
||
|
value += String.fromCharCode(c + 0x20); // lowercase
|
||
|
} else {
|
||
|
value += String.fromCharCode(c);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
return { i };
|
||
|
}
|
||
|
|
||
|
function extractCharacterEncodingFromMeta(string) {
|
||
|
let position = 0;
|
||
|
|
||
|
while (true) {
|
||
|
let subPosition = string.substring(position).search(/charset/i);
|
||
|
|
||
|
if (subPosition === -1) {
|
||
|
return null;
|
||
|
}
|
||
|
subPosition += "charset".length;
|
||
|
|
||
|
while (isSpaceCharacter(string[subPosition].charCodeAt(0))) {
|
||
|
++subPosition;
|
||
|
}
|
||
|
|
||
|
if (string[subPosition] !== "=") {
|
||
|
position = subPosition - 1;
|
||
|
continue;
|
||
|
}
|
||
|
|
||
|
++subPosition;
|
||
|
|
||
|
while (isSpaceCharacter(string[subPosition].charCodeAt(0))) {
|
||
|
++subPosition;
|
||
|
}
|
||
|
|
||
|
position = subPosition;
|
||
|
break;
|
||
|
}
|
||
|
|
||
|
if (string[position] === "\"" || string[position] === "'") {
|
||
|
const nextIndex = string.indexOf(string[position], position + 1);
|
||
|
|
||
|
if (nextIndex !== -1) {
|
||
|
return whatwgEncoding.labelToName(string.substring(position + 1, nextIndex));
|
||
|
}
|
||
|
|
||
|
// It is an unmatched quotation mark
|
||
|
return null;
|
||
|
}
|
||
|
|
||
|
if (string.length === position + 1) {
|
||
|
return null;
|
||
|
}
|
||
|
|
||
|
let end = string.substring(position + 1).search(/\x09|\x0A|\x0C|\x0D|\x20|;/);
|
||
|
if (end === -1) {
|
||
|
end = string.length;
|
||
|
}
|
||
|
return whatwgEncoding.labelToName(string.substring(position, end));
|
||
|
}
|
||
|
|
||
|
function isSpaceCharacter(c) {
|
||
|
return c === 0x09 || c === 0x0A || c === 0x0C || c === 0x0D || c === 0x20;
|
||
|
}
|