Testing various characters

[golden_labels]

I’m just testing some characters in relation to the recent database issue, away from the main thread to not cause noise.

My original post: ✗
At-sign, pound, euro (U+0040, U+00A3, U+20AC): ✗
Euro, pound, at-sign (different order): ✓/✗
£ €: ✓
Pound, at: ✓/✗
≥ …: ✓
greater-eq ellipsis at: ✓/✗

So I could post euro-pound-at sequence below, but I can neither send it in this post nor in the next message. And then I could again.

I could also send pound-at below, but I could not put it in this post.

Note that ✓ and ✗ pass with no issues (U+2713, U+2717). I now edited the pound-at post and added with ✓✗ too, and that still works. So these are not the factor and it’s not “exotic” codepoints that cause the issue.

Here ≥ … work with no problems, but I had to remove them in another post. I can even add the @ below. But if I try to add it here, it does not work. Except that I just did in the previous sentence and it passed.

At this point in this single post I collected all the mentioned/spotted codepoints. And it passes, despite it fails when I try to send them in other positions.

[golden_labels]

Euro, pound, at-sign (different order)

€ £ @

[golden_labels]

Euro, pound, at-sign, again

€ £ @

aaaaaaaaaaand

[golden_labels]

Euro and pound only:

£ €

aaaand

[golden_labels]

Pound and at only:

£ @

aaaand (✓ ✗ ?)

[golden_labels]

Some more codepoints that were in my post that didn’t go through:

≥ …

aaaand

[golden_labels]

Adding at-sign to the mix:

≥ … @

aaaand

[Nominal Animal]

So I could post euro-pound-at sequence below, but I can neither send it in this post nor in the next message.

Could it be interference from the Mentions facility/processing filter?

One way to defuse that is to add say [i][/i] after the at sign, similar to how you can emit \$ without triggering the inline-MathJax conversion.

[golden_labels]

The mentions feature has been present last month too, and in the years before. Why would suddenly so many people be affected by it and why would it be reported as a database error?

I’m not dismissing the idea. The questions are meant to be investigative, not critical.

edit: all these three fail (fullwidth used to circumvent special treatment and the bug):
＠［ｉ］［／ｉ］ £ €
［ｉ］＠［／ｉ］ £ €
＠ ［ｉ］£ €［／ｉ］

It also fails, if the entire at-pound-euro sequence is placed inside the code tags, which should disable any special processing.

[Nominal Animal]

The questions are meant to be investigative, not critical.

Yes, I understood them as such also.

The @ character is not a "special character" in PHP or SQL queries.  It is only "special" with respect to Mentions when SMF BBcode is processed.  I suspect a bug in the related PHP code, triggered by an update in the PHP interpreter or change in server/PHP configuration.  Even the Database error fits, if one looks at 2.1beta1/Sources/Mentions.php:class Mentions (closest version to 2.0.19 I could find online).

[golden_labels]

That’s purely speculative. But let’s assume the following scenario:

• Some table definition has no explicit encoding set on text fields.
• There is a bug in code. It improperly detects mentions, but since the invalid detections are not actual usernames, they’re silently ignored and the bug lives unnoticed.
• The RDBMS has been recently updated to a newer version, which changes the fallback/default encoding used.

I can imagine that after the update the invalid usernames cause an encoding error, instead of being silently ignored.

That would be a very nice mandelbug and shroedinbug at the same time.