Dear Uwe,
Le 27/10/2015 00:40, Uwe Stöhr a écrit :
Am 26.10.2015 um 17:21 schrieb Guillaume Munch:
The user reported two valid use cases that are exceptions to this
behaviour: \lstinline and \usepackage{bigfoot}. This used to work before
⇒ regression.
As I wrote, I don't agree. We had a valid bug report where a user put a
listings inset in a footnote and the result was uncompilable.
I would be more convinced by your message if:
* you referred to #9321 as it is: a user reporting a crash,
* you were able to explain the cause of this crash,
* that point was not already raised in my message.
Therefore
this was forbidden and even backported to LyX 2.1.x
That nobody objected to your patch back then does not sound like an
argument to me.
If you want a proper fix, then we must automatically add the "inline"
option if a listings inset is placed into a footnote or margin note.
I like your suggestion, but now see my other message.
That one can use listings in footnote nevertheless with special settings
or by adding preamble code doesn't matter in my opinion. The default
placement does not compile and average users obviously run into problems.
Imagine you write a thesis and your document becomes uncompilable. You
don't know why because the LaTeX error message is mystery for you since
you use LyX to hide LaTeX commands from you. That could suck a lot!
Those who know special settings and even preamble code to get listings
in footnotes, can add \footnote{ as ERT before the listing and } behind
it. That is not much work and you still have LyX's full listings features.
So the question is who we address as user? In my opinion we should
address average users and they use in most cases default settings since
they are no experts. If this breaks compilation we should not allow this.
I see several issues with your approach.
You keep referring to "average users" vs. "expert users" but this
distinction seems quite idealised. Anybody can copy-paste LaTeX code
from the internet. On the other hand, from past experience with using
ERT for a specialised package, this solution is too cumbersome to be
advocated as long-term.
You propose to apply an idealised principle pushed to the extreme. A
crash is "fixed" by curing the symptom. That a user took the time to
complain about a regression does not count in the balance. ERT becomes a
permanent solution rather than a workaround, in the name of principles.
And to me it looks like it is missing the point, because the compilation
argument is the only one which I did not object to; I even suggested
that it was a separate issue.
Sincerely,
Guillaume
