回應可能涉及他人隱私與回復者的著作權、被遺忘權，因此未必適合公開。
而且把回應分別讀取應該也有利於減少讀取時間…

An element should be removed only after all of what it would do is all finished,
so while(array_pop($stack)) is not a good idea.
Maybe combined with #27.

假設使用者看得到某篇非公開文章，而該文章的連結被貼在某篇公開文章的回應中，那麼抓取該回應時，會不會因為使用者有權限看得到，而「不小心」預覽了非公開文章？

私人備份時無妨，但備份後若有公開需求則須留意。

Figure out a way to run codes only after both of them are ready.
index.php and dt/template.html have not guaranteed main run after FB.apiwt is defined.
See Facebook Docs.

Not only pages and posts, but also comments and their comments.

mongo is going to be deprecated, and mongodb is suggested.

Also add permission selection.
Official reference

added but don't know what to display...

For example, photo node 365599393597218 seems to be a PNG instead of JPEG. But such information is lost for index.html to render the full filename to show the downloaded photo.
And prepare for video files....

While running a update crawling, we shall skip re-crawl the comments of one node if the updated_time of the node hasn't being changed.

利於顯示活動和社團的密集發文。
只要把 filter 改成 limitTo 即可，也不影響 search 的運作，應該不難。

May not be so important, and notice that public groups usually have much more messages and spams, which cost memory and time.

related with #15.

Add some fields such as last_crawled_time, status (for deleted nodes), and _id, in order to keep crawled data as raw as possible.

例如把 message 中的文字用 message_tags 來加上連結。
函數已經寫好了，是 js/utility.js 的 combine_tags ，問題是如何在 AngularJS 中使用。

For example,
1494066394244026_1538635866453745 (node type "post") and 1538635866453745 (node type "video") are sharing the equivalent set of comments (but with different order), so we don't have to crawl them both.

But what about output?

Maybe to "/metadata/v2.5.json" or other meaningful path.

例如是否要下載圖片、回應
全部（重）抓，抑或只抓新的，而所謂「新的」是指 updated_time 有變動，還是 created_time 在某個區間的？
還有要否檢查節點是否仍存
related with #10, #14, #15, and #22.

Deprecate window.fbAsyncInit

Use display: table-cell; of CSS to implement single-line term-description pair instead.
It's so sad that only DT and DD are allowed in DL.

Two possible solutions:

Use several DL for each DT-DD pair.

Cons: Not easy for machines to know they are actually one list, which means the terms have some same properties.

<div style="display: table;">
    <dl style="display: table-row;">
        <dt style="display: table-cell;">term</dt>
        <dd style="display: table-cell;">description</dd>
    </dl>
    <dl style="display: table-row;">
        <dt style="display: table-cell;">longer term</dt>
        <dd style="display: table-cell;">description</dd>
    </dl>
</div>

DL has its style display: table-row is for a table-like display. Otherwise the DTs of different DL would have different widths.
A wrapper is needed to separate the content from other list (if exists).

Use LI to wrap each term-description pair.

Just like the above, but change the wrapper to UL, DL to LI, DT and DD to DIV (or other wrapper).
Pros: You can see them as a list still.
Cons: Not easy for machines to know the term-description relation. (I think this is much more terrible than the other solution, but Facebook use this.)

持續抓取時，若要新增另一個要抓的東西進入排程（這個「新增」似乎用 queue的概念較妥），不宜影響 crawler.php 的 sleep 狀態，所以「把要抓的東西放進 session 」這件事應該放在另一個頁面。
這樣主頁面持續讀取 crawler.php 就好。
新頁面也許可叫做 queueAppend.php 。

硬碟容量反而不是考量重點，而是如何決定儲存路徑。