They preclude ASCII, but not Unicode diagrams.

Unicode block characters have constant width.

Unicode also contains fixed-width equivalents to ASCII letters: ａｂｃ. Just add 0xFEE0 to the code point.

And 0x3000 is an ideographic space, with the same width.

    +---------+
    |         |                        +--------------+
    |   NFS   |--+                     |              |
    |         |  |                 +-->|   CacheFS    |
    +---------+  |   +----------+  |   |  /dev/hda5   |
                 |   |          |  |   +--------------+

turns into

    ＋－－－－－－－－－＋
    ｜         ｜                        ＋－－－－－－－－－－－－－－＋
    ｜   ＮＦＳ   ｜－－＋                     ｜              ｜
    ｜         ｜  ｜                 ＋－－＞｜   ＣａｃｈｅＦＳ    ｜
    ＋－－－－－－－－－＋  ｜   ＋－－－－－－－－－－＋  ｜   ｜  ／ｄｅｖ／ｈｄａ５   ｜
                 ｜   ｜          ｜  ｜   ＋－－－－－－－－－－－－－－＋

EDIT: which would have looked much better had HN not replaced full spaces with the narrow kind.

Those diagrams might look nice, but are hard to edit (for your average user without a CJK input method) and hard to copy things from (you can't get the string /dev/hda5 from your example, you'll get something Linux won't do anything meaningful with.)

I don't think diagrams are meant to be copy-pasted from a lot, so I'm not worried about that.

For editing, I'd just use a conversion tool, and edit it in a fixed-width editor rather than directly in code. Yes, it's an extra tool, but I don't think it's that outlandish to edit pictures in a tool for editing pictures.

My experience is that I get annoyed when text isn't copyable whether or not someone means for it to be copyable.

... but issues would never happen outside of HN, so this is all cool.