- Dillo-dev - lists.mailman3.com

what crashes remain?

by jcid＠dillo.org

On Sun, Sep 28, 2008 at 12:52:47AM +0000, corvid wrote: > Johannes wrote: > > On Sat, Sep 27, 2008 at 05:29:28PM +0000, corvid wrote: > > > I can't think of any other crashes, but I don't remember whether > > > Johannes still gets bus errors. > > > > No, they got fixed with > > http://lists.auriga.wearlab.de/pipermail/dillo-dev/2008-July/004649.html > > > > The only issues I'm seeing are the 128 tabs limit from fltk and the > > redraw storms related to the MenuTabPager. > > Playing with the file menu, I see that Alt-W, Alt-E, Alt-X, Alt-C > all segfault. I thought somebody said something (that I didn't pay > attention to) about the file menu since tabs were added, but if so, > I can't find that mail. Please check the patch in CVS. FLTK2 has a bug in that it regards the first menu item letter as a shortcut. Note: I tried with invisible shortcuts doing nothing, removing shortcuts and all that stuff, until I found this way to workaround the problem. Note2: Alt-W and Alt-X are accelerators, so they're OK. AFAIS there's no more segfault here. -- Cheers Jorge.-

Sept. 29, 2008

patch: show_msg=NO disable debug messages

by onepoint＠starurchin.org

On Thu, Sep 25, 2008 at 02:03:39PM -0400, Jorge Arellano Cid wrote: > In other words, for me, its OK to get rid of DEBUG_MSG unless > somebody else finds good use for it. I usually add my own debugging statements rather than use DEBUG_MSG . Maybe that says more about my ego than anything else :-) but FWIW I wouldn't miss DEBUG_MSG either. Regards, Jeremy Henty

Sept. 29, 2008

patch: changeLinkColor

by jcid＠dillo.org

On Sun, Sep 28, 2008 at 04:40:53PM -0400, Jorge Arellano Cid wrote: > On Sun, Sep 28, 2008 at 09:20:17PM +0200, Johannes Hofmann wrote: > > On Sun, Sep 28, 2008 at 02:54:56PM -0400, Jorge Arellano Cid wrote: > > > On Sun, Sep 28, 2008 at 05:55:10PM +0000, corvid wrote: > > > > Encouraged by the idea that I am going in the right direction, I have > > > > made a new one with spaceStyle: > > > > > > Funny how we were working on the same patch a the same time! > > > > > > I had an ancient version, and gave it a review. > > > The link search is linear inside textblock. A linux gazette page > > > showed more than 33000 words to color a link. On bigger pages it may be worst. > > > It depends on the structure. For instance our bookmarks uses tables > > > and the link is usually found in less than 10 words. > > > > > > The patch in CVS reduce the 35000 steps to 1500 or so (the > > > sorted link list is sparse, so there's a penalty for looking > > > for an actual link to compare to). > > > > > > > Not sure I understand this optimization. > > It is assuming that link numbers are always increasing, right? > > Right. > > > Can we assume this in dw/*? > > I checked html.cc and x_link is crescent. Not sure about the rest. > > > src/html.cc seems to be implemented currently in way that this is > > true. But perhaps we could/should reuse link numbers for equal > > destination urls in the future? > > In that case it'd need a revision. > > > Or someone might just choose hashes of the url string as link > > number. > > > > BTW. with both implementations I now see issues with this example: > > > > <html> > > <body> > > <a href="http://www.dillo.org">dillo</a> > > <a href="http://slashdot.org">slashdot</a> > > <a href="http://www.dillo.org">dillo</a> > > </body> > > </html> > > > > When opening the first link in a new window, it doesn't change color > > in the original one. > > When opening it in a tab, the first one changes color. After a > > reload also the second one changes color. > > Am I missing something? > > Yes, the link color change is meant as a visual cue for the user > to know the click is being processed. We could leave the link alone > and change the mouse cursor instead. > > The link not changing color when openning it with the menu > is IMO a bug, I'm looking for... It seems that carrying around widget, link and html->visited_color could do the trick. Maybe it's not worth the effort as the new window is enough visual feedback, and also the new tab... Comments? -- Cheers Jorge.-

Sept. 28, 2008

patch: changeLinkColor

by jcid＠dillo.org

On Sun, Sep 28, 2008 at 09:20:17PM +0200, Johannes Hofmann wrote: > On Sun, Sep 28, 2008 at 02:54:56PM -0400, Jorge Arellano Cid wrote: > > On Sun, Sep 28, 2008 at 05:55:10PM +0000, corvid wrote: > > > Encouraged by the idea that I am going in the right direction, I have > > > made a new one with spaceStyle: > > > > Funny how we were working on the same patch a the same time! > > > > I had an ancient version, and gave it a review. > > The link search is linear inside textblock. A linux gazette page > > showed more than 33000 words to color a link. On bigger pages it may be worst. > > It depends on the structure. For instance our bookmarks uses tables > > and the link is usually found in less than 10 words. > > > > The patch in CVS reduce the 35000 steps to 1500 or so (the > > sorted link list is sparse, so there's a penalty for looking > > for an actual link to compare to). > > > > Not sure I understand this optimization. > It is assuming that link numbers are always increasing, right? Right. > Can we assume this in dw/*? I checked html.cc and x_link is crescent. Not sure about the rest. > src/html.cc seems to be implemented currently in way that this is > true. But perhaps we could/should reuse link numbers for equal > destination urls in the future? In that case it'd need a revision. > Or someone might just choose hashes of the url string as link > number. > > BTW. with both implementations I now see issues with this example: > > <html> > <body> > <a href="http://www.dillo.org">dillo</a> > <a href="http://slashdot.org">slashdot</a> > <a href="http://www.dillo.org">dillo</a> > </body> > </html> > > When opening the first link in a new window, it doesn't change color > in the original one. > When opening it in a tab, the first one changes color. After a > reload also the second one changes color. > Am I missing something? Yes, the link color change is meant as a visual cue for the user to know the click is being processed. We could leave the link alone and change the mouse cursor instead. The link not changing color when openning it with the menu is IMO a bug, I'm looking for... BTW, the linear search patch is much simpler and as link clicking is not a performance critical task it could be left implemented in O(n). -- Cheers Jorge.-

Sept. 28, 2008

patch: changeLinkColor

by corvid＠lavabit.com

Johannes wrote: > BTW. with both implementations I now see issues with this example: > > <html> > <body> > <a href="http://www.dillo.org">dillo</a> > <a href="http://slashdot.org">slashdot</a> > <a href="http://www.dillo.org">dillo</a> > </body> > </html> > > When opening the first link in a new window, it doesn't change color > in the original one. > When opening it in a tab, the first one changes color. After a > reload also the second one changes color. > Am I missing something? Middle click changes the color for me whether it opens a tab or a window.

Sept. 28, 2008

patch: changeLinkColor

by Johannes.Hofmann＠gmx.de

On Sun, Sep 28, 2008 at 07:42:14PM +0000, corvid wrote: > Jorge wrote: > > On Sun, Sep 28, 2008 at 05:55:10PM +0000, corvid wrote: > > > Encouraged by the idea that I am going in the right direction, I have > > > made a new one with spaceStyle: > > > > Funny how we were working on the same patch a the same time! > > > > I had an ancient version, and gave it a review. > > The link search is linear inside textblock. A linux gazette page > > showed more than 33000 words to color a link. On bigger pages it may be worst. > > It depends on the structure. For instance our bookmarks uses tables > > and the link is usually found in less than 10 words. > > > > The patch in CVS reduce the 35000 steps to 1500 or so (the > > sorted link list is sparse, so there's a penalty for looking > > for an actual link to compare to). > > > > Please apply the spaceStyle part, and send it to me. > > Do we know that all of the (TEXT -- putting WIDGET aside for the moment) > contents of a link will have the same style? First likely example I could > come up with is how about in a utopian future dillo where tooltips are > implemented and there's an ABBR inside? Oh right. Just think about: <html> <body> <a href="http://slashdot.org">slash<b>dot</b></a> </body> </html> After clicking, the dot is no longer bold. Cheers, Johannes

Sept. 28, 2008

patch: changeLinkColor

by Johannes.Hofmann＠gmx.de

On Sun, Sep 28, 2008 at 02:54:56PM -0400, Jorge Arellano Cid wrote: > On Sun, Sep 28, 2008 at 05:55:10PM +0000, corvid wrote: > > Encouraged by the idea that I am going in the right direction, I have > > made a new one with spaceStyle: > > Funny how we were working on the same patch a the same time! > > I had an ancient version, and gave it a review. > The link search is linear inside textblock. A linux gazette page > showed more than 33000 words to color a link. On bigger pages it may be worst. > It depends on the structure. For instance our bookmarks uses tables > and the link is usually found in less than 10 words. > > The patch in CVS reduce the 35000 steps to 1500 or so (the > sorted link list is sparse, so there's a penalty for looking > for an actual link to compare to). > Not sure I understand this optimization. It is assuming that link numbers are always increasing, right? Can we assume this in dw/*? src/html.cc seems to be implemented currently in way that this is true. But perhaps we could/should reuse link numbers for equal destination urls in the future? Or someone might just choose hashes of the url string as link number. BTW. with both implementations I now see issues with this example: <html> <body> <a href="http://www.dillo.org">dillo</a> <a href="http://slashdot.org">slashdot</a> <a href="http://www.dillo.org">dillo</a> </body> </html> When opening the first link in a new window, it doesn't change color in the original one. When opening it in a tab, the first one changes color. After a reload also the second one changes color. Am I missing something? Cheers, Johannes

Sept. 28, 2008

what crashes remain?

by corvid＠lavabit.com

Jorge wrote: > On Sat, Sep 27, 2008 at 11:15:42PM +0000, corvid wrote: > > Jorge wrote: > > > On Sat, Sep 27, 2008 at 05:29:28PM +0000, corvid wrote: > > > > It's still possible to crash dillo if you click on a remote link and > > > > quickly click on some images to load them. Doesn't crash every time, > > > > but typically within four tries or so. I had assumed it was the same > > > > thing as whatever was wrong with the menu crash, so I'd ignored it. > > > > > > > > If I stick > > > > a_Bw_stop_clients(bw, BW_Img); > > > > in the DilloHtml destructor, I can no longer get it to crash, but > > > > I randomly got a "Cache_process_queue Caught busy!!!" while browsing > > > > earlier today, so there's something wrong with just doing that. > > > > (Assuming it's related; I'm not in the habit of getting that MSG.) > > > > > > That message means big trouble: the event loop stepped over itself... > > > > > > I've tried several different ways but can't get to crash it. > > > > > > Does the attached patch help? > > > > Seems to be working well so far. > > Is there any danger in adding something like > > a_Nav_cancel_expect to a_UIcmd_stop()? > > What's the idea behind it? > > If you want to get rid of empty cache entries from > unsuccesful http connections, it's more complex than that. > BTW, I plan to do that after the release. I'm on page A. I try to go to page B, which seems to be taking a long time to think about resolving or contacting the host or whatever. I hit stop and go back to reading and want to continue loading images. Or if DNS couldn't resolve the hostname, I could hit stop to re-enable loading images.

Sept. 28, 2008

what crashes remain?

by jcid＠dillo.org

On Sat, Sep 27, 2008 at 11:15:42PM +0000, corvid wrote: > Jorge wrote: > > On Sat, Sep 27, 2008 at 05:29:28PM +0000, corvid wrote: > > > It's still possible to crash dillo if you click on a remote link and > > > quickly click on some images to load them. Doesn't crash every time, > > > but typically within four tries or so. I had assumed it was the same > > > thing as whatever was wrong with the menu crash, so I'd ignored it. > > > > > > If I stick > > > a_Bw_stop_clients(bw, BW_Img); > > > in the DilloHtml destructor, I can no longer get it to crash, but > > > I randomly got a "Cache_process_queue Caught busy!!!" while browsing > > > earlier today, so there's something wrong with just doing that. > > > (Assuming it's related; I'm not in the habit of getting that MSG.) > > > > That message means big trouble: the event loop stepped over itself... > > > > I've tried several different ways but can't get to crash it. > > > > Does the attached patch help? > > Seems to be working well so far. > Is there any danger in adding something like > a_Nav_cancel_expect to a_UIcmd_stop()? What's the idea behind it? If you want to get rid of empty cache entries from unsuccesful http connections, it's more complex than that. BTW, I plan to do that after the release. > PS Stop should have a keyboard shortcut, I think. > I long had a reflex of hitting ESC from the netscape/mozilla > days, but of course ESC would be inconsistent for dillo. Could be. Maybe Ctrl-P. Although Stop is not yet doing exactly what it should. -- Cheers Jorge.-

Sept. 28, 2008

patch: changeLinkColor

by corvid＠lavabit.com

I wrote: > The registerName is because html.cc needs to have that working in > order to call changeLinkColor. By the way, in addition to html.cc, instanceOf is used in Textblock::addParbreak(). Perhaps we will see some effect from this.

Sept. 28, 2008