April 2009 - Dillo-dev - lists.mailman3.com

Segfault on sites requiring authentication

by tim.nieradzik＠gmx.de

> I'm not seeing this crash. Is it going through the for loop and > getting set to NULL in Client_bw = ClientWeb->bw;, or is it > NULL because there weren't any clients and it didn't execute that code? Yes, it is going through the loop. Probably comparing Client->Url with entry->Url fails. As a result, ClientWeb and Client_bw do not get set. Have you also tried reloading the page? Sometimes it does not get loaded properly here. Anyway, it is odd that Dillo does not crash for you. Are you using the latest trunk? --Tim

April 1, 2009

[PATCH] startup window

by Johannes.Hofmann＠gmx.de

On Wed, Apr 01, 2009 at 04:49:44PM +0000, corvid wrote: > Thomas wrote: > > Am Wed, 1 Apr 2009 09:47:11 +0200 > > schrieb Hofmann Johannes <Johannes.Hofmann(a)gmx.de>: > > > > > Hi, > > > attached patch adds a proper startup screen for dillo. > > > > Oh, is dillo slow enough now that it needs a splash screen distracting the user while waiting for startup? > > I sort of like the "instant startup" behaviour. A splash startup screen reminds you of the startup _not_ being instant. > > Especially opening an extra window wastes CPU cycles, even in X11, that could be used on doing the actual work... > > Of course I'll have to check out a current dillo2 to see how eminent the issue really is. But when we reach the startup speed of e.g. the GIMP with it's endless loading of plugins etc., one can safely declare dillo "just another bloated browser". > > I don't want to be offensive, just voice my concern... > > I agree with Thomas here. > me too - check the date :)

April 1, 2009

[PATCH] startup window

by corvid＠lavabit.com

Thomas wrote: > Am Wed, 1 Apr 2009 09:47:11 +0200 > schrieb Hofmann Johannes <Johannes.Hofmann(a)gmx.de>: > > > Hi, > > attached patch adds a proper startup screen for dillo. > > Oh, is dillo slow enough now that it needs a splash screen distracting the user while waiting for startup? > I sort of like the "instant startup" behaviour. A splash startup screen reminds you of the startup _not_ being instant. > Especially opening an extra window wastes CPU cycles, even in X11, that could be used on doing the actual work... > Of course I'll have to check out a current dillo2 to see how eminent the issue really is. But when we reach the startup speed of e.g. the GIMP with it's endless loading of plugins etc., one can safely declare dillo "just another bloated browser". > I don't want to be offensive, just voice my concern... I agree with Thomas here.

April 1, 2009

Segfault on sites requiring authentication

by corvid＠lavabit.com

Tim wrote: > Whenever I visit http://puppyisos.org/isos (puppy / linux), Dillo > segfaults. The reason is that in some cases the variable Client_bw > is NULL. However, some functions depend on it and access its member > variables. > > Client_bw is initialized in Cache_process_queue() (src/cache.c). > When authentication is required, this variable is passed over to > Cache_auth_entry() which sets a timeout. Its callback > (Cache_auth_callback()) opens up a dialog asking for the credentials > and then calls a_Nav_reload() with the Client_bw as its first > parameter. Finally in a_Nav_cancel_expect() Client_bw's member > variable nav_expecting is used and as bw is NULL, Dillo segfaults. I'm not seeing this crash. Is it going through the for loop and getting set to NULL in Client_bw = ClientWeb->bw;, or is it NULL because there weren't any clients and it didn't execute that code? > I came up with a small patch that removes Cache_auth_callback() and > Cache_auth_entry() as I have not seen any advantages in delaying the > authentication popup. The next thing I did was to show the dialog on > condition that Client_bw is not NULL and that we have finished the > transfer (At least I hope CA_Close is the proper flag for doing so). > > There were various unnecessary checks in src/nav.c. They compared bw > with NULL. In my opinion the affected functions (a_Nav_expect_done, > a_Nav_push and a_Nav_repush) should never be called if you cannot be > sure that bw is not NUL. What is your view on this subject? > > Unfortunately, my approach does not solve the original problem that > caused Client_bw being NULL. Furthermore, the patch introduces a new > bug: The dialog pops up when reloading the site but after having > entered the correct credentials or having pressed the ``Cancel'' > button, it appears again. Any ideas? > > While digging deeper into the code, I noticed that we make heavy use > of timers for popups and reloading pages? Is there any special > reason for it? If the calling code needs to go back to whatever it was doing rather than waiting around for the user to interact with a popup, a timer is used.

April 1, 2009

[PATCH] startup window

by thomas-forum＠orgis.org

Am Wed, 1 Apr 2009 09:47:11 +0200 schrieb Hofmann Johannes <Johannes.Hofmann(a)gmx.de>: > Hi, > attached patch adds a proper startup screen for dillo. Oh, is dillo slow enough now that it needs a splash screen distracting the user while waiting for startup? I sort of like the "instant startup" behaviour. A splash startup screen reminds you of the startup _not_ being instant. Especially opening an extra window wastes CPU cycles, even in X11, that could be used on doing the actual work... Of course I'll have to check out a current dillo2 to see how eminent the issue really is. But when we reach the startup speed of e.g. the GIMP with it's endless loading of plugins etc., one can safely declare dillo "just another bloated browser". I don't want to be offensive, just voice my concern... Alrighty then, Thomas.

April 1, 2009

[PATCH] startup window

by Johannes.Hofmann＠gmx.de

Hi, attached patch adds a proper startup screen for dillo. Cheers, Johannes

April 1, 2009

Segfault on sites requiring authentication

by tim.nieradzik＠gmx.de

Whenever I visit http://puppyisos.org/isos (puppy / linux), Dillo segfaults. The reason is that in some cases the variable Client_bw is NULL. However, some functions depend on it and access its member variables. Client_bw is initialized in Cache_process_queue() (src/cache.c). When authentication is required, this variable is passed over to Cache_auth_entry() which sets a timeout. Its callback (Cache_auth_callback()) opens up a dialog asking for the credentials and then calls a_Nav_reload() with the Client_bw as its first parameter. Finally in a_Nav_cancel_expect() Client_bw's member variable nav_expecting is used and as bw is NULL, Dillo segfaults. I came up with a small patch that removes Cache_auth_callback() and Cache_auth_entry() as I have not seen any advantages in delaying the authentication popup. The next thing I did was to show the dialog on condition that Client_bw is not NULL and that we have finished the transfer (At least I hope CA_Close is the proper flag for doing so). There were various unnecessary checks in src/nav.c. They compared bw with NULL. In my opinion the affected functions (a_Nav_expect_done, a_Nav_push and a_Nav_repush) should never be called if you cannot be sure that bw is not NUL. What is your view on this subject? Unfortunately, my approach does not solve the original problem that caused Client_bw being NULL. Furthermore, the patch introduces a new bug: The dialog pops up when reloading the site but after having entered the correct credentials or having pressed the ``Cancel'' button, it appears again. Any ideas? While digging deeper into the code, I noticed that we make heavy use of timers for popups and reloading pages? Is there any special reason for it? --Tim

April 1, 2009