Adding Daisy as owner.

I could not repro this issue on 51.0.2702.0 (8088.0.0 dev-channel link test)

I'm still seeing it in M50 (beta), but I haven't tried M51 yet.

Re#4: could you attach your Chrome version please? I'll test on M50.
Since M49 stable has shipped, change the label to M50.

Version 50.0.2661.57 beta (64-bit)
Platform 7978.36.0 (Official Build) beta-channel samus
Firmware Google_Samus.6300.174.0

I can repro it on M50, but it seems that it has been fixed in M51. 

I did some comparison of the fullwidth punctuations' rendering on M50 and M51 (see attached images). I suspect there might be some problem with the bundled fonts in M50. 
+msw@, +jshin@, in case they know the reason.

Deleted: M50.png 80.8 KB

	M50.png 80.8 KB View Download

Deleted: M51.png 48.8 KB

	M51.png 48.8 KB View Download

Sorry, I don't have any clue; maybe some MD work is touching fonts?

msw@, what is MD?

Material Design.

Not likely to be a font change (although I haven't checked anything, yet), but it's more likely to be blink.

Xdai@, how/where did you take PNG images in comment 9? 

Btw, why is the view restricted to Google? There's nothing confidential here? I would lift the restriction, but there is no view-restrcted-google label to drop that used to be present for such a bug.

Looks like they are screenshots of a test web page. Does m50 'transform' full width characters to their ASCII counterpart if the first font in the font stack does not cover them? 

R50/M50 branch was cut on Feb 26. Noto fonts were updated on March 16. So they're not in M50.
Moreover, this update didn't include CJK. The last Noto Sans CJK update was last September (to 1.004). 
 
The font update before that was on November 4 for croscore fonts (Arimo for Arial, Tinos for Times New Roman, etc). Because there's a regression in hmtx table, this update was merged to R46 and R47 as well as R48. 

So, the "regression" found in R49 and R50 can't be explained by a font change. 
 

With Chrome OS R49 (English locale), I confirmed what I suspected based on the screenshots in comment 9. (I used 
https://en.wikipedia.org/wiki/Halfwidth_and_fullwidth_forms and DOM Inspector - Computed - Rendered font). 

For 'font-family: Roboto;', full-width ASCII (U+FFxx) block characters are rendered with Roboto. 

For 'font-family: Arial;" or 'font-family: sans-serif' (with the default font preerence settings), full-width ASCII (U+FFxx) is rendered with 'Arimo' (Arial-equivalent). 

Roboto does NOT cover those characters (U+FFxx). How can it render U+FFxx? The same can be said of Arial/Arimo. 

In M49/M50, Blink might have transformed U+FFxx characters to the corresponding ASCII code points if the first font (??) in the font stack does not cover U+FFxx characters. 

I can't reproduce the problem on Mac, but it may happen on Linux, too. 

drott@, kojii@: Do you have any hunch? M51/R51 is all right, but M49/R49 and  M50/R50 have this issue.  

This looks like dup of  issue 569997 , which is getting quite hot. It'd be great if you could have a look.

Yes, I think it's the same issue as  Issue 569997 . And I think we do need to cherry-pick the fix in M50 since it's really annoying (based on my own experience, sometime it's unacceptable to use the half-width characters.) jshin@, are you the right person to do the cherry-pick?

Thanks for the pointer. Gee... HB was doing NFKD ...

I'll cherry-pick that patch. 
BTW, can you please remove 'Restrict-View-Google' from this bug? 

Why don't I see that label? I have no idea. 

Me neither... I guess it's a unidirectional label?

Got a tip from a monorail team member :-)