[Edbrowse-dev] URL assert

[Edbrowse-dev] URL assert

[Karl Dahlke]

> Are any other classes effected?

No. I checked.
One of those cut & paste things,
I copied the window class struct to build the url class,
and the global flag went along for the ride, and it shouldn't have.
I was happy it was a trivial fix, not a major error in logic or structure.

The new smtp pop3 machinery is in place, using the curl library,
which we were already using for http and ftp anyways.
Thanks to Chris.
Saves about 1200 lines of code, is easier to read,
is maintained by someone else,
and provides some new features that I don't have to reinvent.
ipv6 mail servers, various authentication algorithms, etc,
but I wonder if we haven't lost the ability to cc,
rather than a straight recipient.
I just thought of that now, as I write this. Hmm.
I suppose I could live without cc, but it's nice sometimes.
Chris what do you think here?

Another thing we save is 52 messages, in messages.c, that aren't needed any more.
I changed these to NotUsed1 through NotUsed52 in messages.h.
This was easier than deleting them, and deleting, perfectly,
the corresponding strings in all the languages we support.
So if you need a new message in the future, glom onto one of the NotUsed
messages and take it over, in its position.
I guess we lost 52 messages because I use the curl messages for everything now,
and there are fewer of them anyways than my home grown software.

On we go.

Karl Dahlke

Re: [Edbrowse-dev] URL assert

[Adam Thompson]

[-- Attachment #1: Type: text/plain, Size: 486 bytes --]

On Thu, Apr 24, 2014 at 02:40:45PM +0000, Karl Dahlke wrote:
> It was easy to fix, just one line of code;
> well easy once I spent 2 hours building the debug environment and tracking it down.
> I had the global flag set in the class spec,
> and that should only be in the window class, not any other class.
> That's it.
> So pull the latest and it should behave.

Ah ok, I did wonder about this but I wasn't sure.
Are any other classes effected?

Thanks for fixing this.

Cheers,
Adam.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]

[Edbrowse-dev] URL assert

[Karl Dahlke]

It was easy to fix, just one line of code;
well easy once I spent 2 hours building the debug environment and tracking it down.
I had the global flag set in the class spec,
and that should only be in the window class, not any other class.
That's it.
So pull the latest and it should behave.

Chris I'll look at your smtp patch later today.

Karl Dahlke

Re: [Edbrowse-dev] URL assert

[Adam Thompson]

[-- Attachment #1: Type: text/plain, Size: 1936 bytes --]

On Wed, Apr 23, 2014 at 02:54:39PM +0000, Karl Dahlke wrote:
> Yes the URL class is a royal pain.
> See jsrt line 537 for all the stuff it's suppose to do.
> Lots of side effects, lots of setters.

Yeah I've had a *brief* look at it and it looks crazy.

> If you find a reproducible pathway to make the assert happen
> I'll track through the code and look for something amiss.
> I don't have a debug version build, but the regular version from Redhat
> I can't get to fail, and I've tried running large blocks
> of my jsrt code in a loop 100,000 times over and it seems to work.
> So I guess I'm just not hitting the right combination.

No, it's just the asserts aren't present in the non-debug builds apparently.
The only way you'll see this is to build a debug build of mozjs (as recommended
for debugging and development on their website).
I suspect there's probably a bug somewhere which is being caught by the debug
build but hasn't (yet) manifested itself.
As for making the assert happen, it's simply a case of loading jsrt (or
seemingly anything) with js enabled on my build of mozjs.

> I have no changes pending, nothing in work.
> Chris is working on something but it's all over in sendmail.c land.
> So if you do have time and want to investigate, the *cpp files are quiescent.
> If other things are going on in your world I understand;
> maybe you can just make the assert happen and I'll take it from there.

I had a *quick* go at debugging this yesterday but I really don't have time to
go through the mozjs code to try and understand why we're breaking something now.
This is because I'm in the final year of my university course and have a couple
of assignments, one of which being an 8000 plus word report,
due in for next week. After that I've got final exams to revise for but I
*might* have a bit more time to do edbrowse related development after next 
Friday.

Cheers,
Adam.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]

[Edbrowse-dev] URL assert

[Karl Dahlke]

Yes the URL class is a royal pain.
See jsrt line 537 for all the stuff it's suppose to do.
Lots of side effects, lots of setters.

If you find a reproducible pathway to make the assert happen
I'll track through the code and look for something amiss.
I don't have a debug version build, but the regular version from Redhat
I can't get to fail, and I've tried running large blocks
of my jsrt code in a loop 100,000 times over and it seems to work.
So I guess I'm just not hitting the right combination.

I have no changes pending, nothing in work.
Chris is working on something but it's all over in sendmail.c land.
So if you do have time and want to investigate, the *cpp files are quiescent.
If other things are going on in your world I understand;
maybe you can just make the assert happen and I'll take it from there.

Cheers,

Karl Dahlke