Topic: 2 crashes and few remarks

freem wrote:

boost::format errors usually occur when there is a problem with substituting place holders in strings, or if a string contains a weird character that can't be handled (e.g. if something is accidentally escaped).

Oh, it's the boost's enhanced sprintf, then? I think I see how it works (but never used it).

Yes, that's my best guess

I tested this again and I still can't reproduce the crash. I loaded the savegame, selected a castle, moved the castle outside of the screen to the bottom right, then moved the Widelands window outside of the screen on the bottom right.

I've launched a gdb instance on the crash, and the line of the crash is game_preload_packet.cc:112. Now, it no longer happens git commit 48e6c8. I have seen nothing in changelogs indicating a change on that file, I'm running git bisect to know when exactly that crash stopped to happens (because it may just have hidden and not actually fix. Plus, I've got some time).

That line doesn't contain any boost stuff. Can you do a backtrace the next tome you catch this?

I wish there was a way to load a map from command-line, would spare some steps. Or better, a way to script actions, but I know this would be quite hard to code

You can run a scenario through the command line. The work would be coding the scenario though.

https://wl.widelands.org/wiki/RegressionTests/

Busy indexing nil values

Top

Quote

First, I am sorry, my reply may be a little messy, my mind is not really, say... clean, at this hour today.

GunChleoc wrote:

freem wrote:

I tested this again and I still can't reproduce the crash. I loaded the savegame, selected a castle, moved the castle outside of the screen to the bottom right, then moved the Widelands window outside of the screen on the bottom right.

I've launched a gdb instance on the crash, and the line of the crash is game_preload_packet.cc:112. Now, it no longer happens git commit 48e6c8. I have seen nothing in changelogs indicating a change on that file, I'm running git bisect to know when exactly that crash stopped to happens (because it may just have hidden and not actually fix. Plus, I've got some time).

That line doesn't contain any boost stuff. Can you do a backtrace the next tome you catch this?

Sure, I'll do it tomorrow. Also, I was unable to find which commit exactly did fixed the crash, I had a mess of savegames making the game crashing a way or another and the game working fine... It seems there were several commits messing with zip files, which did or not recognize the savegame. That lost me a lot, one commit crashing for a reason and the next one for another one :/. At least, I guess I could speak about savegame changes, improvements or not, in my «journal» on linuxfr :D And, to be honest I am quite tired of eisenbugs. But well, I really like widelands so I intend to hunt that bug. That will be a... hum... payment... for all the fun I had playing it and saying nothing about that

I wish there was a way to load a map from command-line, would spare some steps. Or better, a way to script actions, but I know this would be quite hard to code

You can run a scenario through the command line. The work would be coding the scenario though.

There is nothing in man widelands saying it is possible. And... well, coding a scenario? I'm not against it, since, tbh, I would like to help improving campaigns, but I know 2 things: 1: I'm not a reliable person. I mean, it's hard to have me really involved in a project for a long time. Small contributions are ok to me, really, and I like to help, but... 2: I have no idea about what the storylines where meant to be, which is quite important for a campaign.

Now, I know my weaknesses, so I know that I am only able to punctual contributions: small patches, bug reports, little user help, and other basic ideas. One of which I am actually trying to do.

SirVer wrote:

I have seen nothing in changelogs indicating a change on that file

It could be that the problem was in a wrongly translated string. someting like "{1}" in english was translated to "{12}" in French which might have been fixed in merging the latest translations.

Yep, I have not really read the code, but that kind of errors can happen with printf when you import the format string from another function. Maybe I should try to help you find from where it comes when several of you are on IRC? What are the moments one of you 2 are on IRC the most frequently, so that you could guide me in finding where the bug happens?

I wish there was a way to load a map from command-line

There is --load and --script which makes it possible to run arbitrary commands - including clicking on any buttons and fields.

Huh... I'm... embarrassed. Really... Now that you mention it, there is some information in widelands --help, which I did not even checked, I'm too used to the man command! This might be a bug report (I'll check anew if I can finally access that damn tracker tomorrow without timeouts :/... I guess you won't read bugs on github, right? At least, if I where you, I would not.) It's my fault, so I apologize, but can I ask you which tool you use to generate the manpages? I regularly wonder if it could be possible to generate manpages from the --help option and same for the autocompletion in commandline. Also, I would like to suggest that writing that in a debug page in the wiki would have helped me a lot. Can't say I don't understand why it wasn't there, I prefer to write code than doc.

Anyway, I'll investigate more on that crash, since it seems only I can reproduce it in a consistent way. Just, please, don't take it as a blocker for next release, since new commits fixes it. Knowing the cause would be a good thing, but I really would like widelands to release more often, so don't block for fixed issues :p After all, I really am always eager to play new widelands releases, even if I never says it (without saying it, I'm following your project since several years, and I'm not sure which one is the cleaner, between widelands and openmw). And I'm probably not the only one.

Finally, by "including clicking"... do you mean it is possible to script windows' moves? That would allow me to write a shell script checking that crash so that I could actually do a small but real contribution to widelands. If I could script the crash on the commit I know there is the crash on, it could eventually become an automated test, and then you could integrate it on brz to be checked before push event (at least, I guess it is possible, I know it is at least partially possible with git, so why not with bzr? I'm simply not familiar with it.)

Top

Quote