27616 - Bad mathml causes crash

Status: VERIFIED FIXED

(Core :: MathML, defect, P3)

Description

[Paul Gartside]

This page http://www.maths.ox.ac.uk/~gartside/mozBugs/genfn.xml loads ok.

This page http://www.maths.ox.ac.uk/~gartside/mozBugs/badgenfn.xml causes a

crash on my linux mathml M13 build (due to Dawn Endico).



The problem seems to be line 429 and following.

I have snipped out this section and put in

http://www.maths.ox/~gartside/mozBugs/mathmlBug.xml



Strangely, this does not cause a crash. -But gives a blank screen. (My apologies

fr not pinning this down more\carefully.)

The mathml in badgenfn (produced by tex4ht) is clearly bad, but a full scale

crash is disappointing!

Comment 1

[Paul Gartside]

Minor error (cough!)

The third url is http://www.maths.ox.ac.uk/~gartside/mozBugs/mathmlBug.xml

Comment 2

[rbs]

The problem is that <msup> expects the 'base' and the 'superscript'.

But only the base is provided in the second erroneous <msup>, and the

code encounters a null pointer:



<p>

<math

class="inline"><msub><mi>&lambda;</mi><mrow><mi>&Upsilon;</mi></mrow></msub>



<msup><mrow><mi>&Omega;</mi></mrow></msup>



</math>

</p>



The markup is bad and we don't do validation. But in general, the code

must be updated to check for such null pointers.

Comment 3

[rbs]

Fixed in my tree. I will land the fix in my next check-in.
Rather than crashing, a feedback is now provided to the user.
By default the user will see the Unicode REPLACEMENT CHAR U+FFFD
at the spot where a bad markup was encountered. (If there is no font 
with U+FFFD, the user will see the question mark '?' --outcome of 
the work done for 6585).

Comment 4

[Paul Gartside]

I guess you havent put your fix in, as I still get a crash (Friday 17th build).

But I look forward to testing your fix soon!

Comment 5

[rbs]

No I haven't yet landed the fix(es). I am currently re-working the architecture
of stretchy chars to support arbitrary fonts (TeX, MT Extra, Mathematica, etc). 
The initial code was just a prototype. I am also bringing remaining Shyjan's TeX
rendering rules in line with the overall MathML architecture and the bounding 
metrics APIs. My tree is in a major state of flux as a result of all these
changes. So you might have to wait a little while to see the fix(es). Hopefully, 
M16 will feature a much improved MathML renderer, with the remaining TeX
alignment rules and in which the font problems that we having now will be 
considerably reduced.

Comment 6

[Paul Gartside]

Fixed now. Should this not have its status changed to `resolved'.

Comment 7

[rbs]

Rendering of bad mathml is mis-aligned. The formatting is interrupted abruptly
when bad markup is detected, and the placements are left incomplete. But I am
changing the status to resolved since it doesn't crash anymore.

Comment 8

[Hixie (not reading bugmail)]

VERIFIED using Windows 2000 Mozilla bits 2000090706-MathML-SVG.