doc: header escaping regression

[rubys]

Details here: #21936 (comment); pulled out as a separate issue so that we can discuss the changes and reference it in a pull request.

[rubys]

I'd prefer not to remove the escapes from the markdown. My thoughts are that the markdown should represent what an ideal markdown tool would consume, and our code needs to handle differences between our chosen tool (currently remark, et. al) and what that ideal is. I've provided a pull request to the code that underlies the remark tool, but it has yet to be evaluated: syntax-tree/mdast-util-to-hast#21

I believe that for now adding something like the following early in the unified processing of headers would address the needs for both HTML and JSON generation:

.replace(/\\./g, (match) => match[1])

I'll test that and provide a pull request. If/when the remark tool gets updated, we can look into removing this.

[rubys]

Not related to the regression, but unless I take specific action to prevent it, a side effect of moving .use(html.preprocessText) earlier in the pipeline is that descriptions in the generated JSON will more closely match the descriptions in the generated HTML. An example to illustrate:

--- api0/all.json 2018-08-01 15:27:02.000000000 -0400
+++ api/all.json 2018-08-01 15:32:21.000000000 -0400
@@ -38,7 +38,7 @@
 {
 "textRaw": "Syscalls and man pages",
 "name": "syscalls_and_man_pages",
- "desc": "<p>System calls like open(2) and read(2) define the interface between user programs\nand the underlying operating system. Node.js functions\nwhich simply wrap a syscall,\nlike <a href=\"fs.html#fs_fs_open_path_flags_mode_callback\"><code>fs.open()</code></a>, will document that. The docs link to the corresponding man\npages (short for manual pages) which describe how the syscalls work.</p>\n<p>Most Unix syscalls have Windows equivalents, but behavior may differ on Windows\nrelative to Linux and macOS. For an example of the subtle ways in which it's\nsometimes impossible to replace Unix syscall semantics on Windows, see <a href=\"https://github.com/nodejs/node/issues/4760\">Node\nissue 4760</a>.</p>",
+ "desc": "<p>System calls like <a href=\"http://man7.org/linux/man-pages/man2/open.2.html\"><code>open(2)</code></a> and <a href=\"http://man7.org/linux/man-pages/man2/read.2.html\"><code>read(2)</code></a> define the interface between user programs\nand the underlying operating system. Node.js functions\nwhich simply wrap a syscall,\nlike <a href=\"fs.html#fs_fs_open_path_flags_mode_callback\"><code>fs.open()</code></a>, will document that. The docs link to the corresponding man\npages (short for manual pages) which describe how the syscalls work.</p>\n<p>Most Unix syscalls have Windows equivalents, but behavior may differ on Windows\nrelative to Linux and macOS. For an example of the subtle ways in which it's\nsometimes impossible to replace Unix syscall semantics on Windows, see <a href=\"https://github.com/nodejs/node/issues/4760\">Node\nissue 4760</a>.</p>",
 "type": "misc",
 "displayName": "Syscalls and man pages"
 }

Note that where the JSON description had open(2) before, and <a href=\"http://man7.org/linux/man-pages/man2/open.2.html\"><code>open(2)</code></a> after.

I actually think it is a good thing for the JSON descriptions to more closely match the posted HTML, and will proceed under that assumption unless somebody objects. This is an example of the type of things I wanted to look at after the change was made and the first set of bugs were identified and resolved.

[rubys]

@vsemozhetbyt I'm not understanding what the issue with the generated JSON is:

 {
 "textRaw": "urlSearchParams[Symbol.iterator]()",
 "type": "method",
 "name": "[Symbol.iterator]",
 "signatures": [
 {
 "return": {
 "textRaw": "Returns: {Iterator}",
 "name": "return",
 "type": "Iterator"
 },
 "params": []
 }

What am I missing?

[vsemozhetbyt]

I mean that now JSON is OK. But if I remove escapes from .md files and change RegExps in json.js, the JSON result is different and some fields are missing.