Linked by David Adams on Fri 5th Aug 2011 16:08 UTC
Thread beginning with comment 483866
To view parent comment, click here.
To read all comments associated with this story, please click here.
To view parent comment, click here.
To read all comments associated with this story, please click here.
RE[3]: Two distinct beasts...
by dragos.pop on Mon 8th Aug 2011 14:57
in reply to "RE[2]: Two distinct beasts..."
Of course, on Linux, this would never work, since the general assumption seems to be that having 900 different toolkits is a good thing. So if GUI automation is hard, it's probably because of a piss-poor implementation, or someone hasn't written the tools to do it. But doesn't that work the same way on CLIs? I mean, SOMEONE has to build in pipes and stuff if you really want to do something useful with it ...
Not necessary... it would work, but you have to chose the solution made for the toolkit that your app is build upon. Plus sending key and mouse events might work the same on all toolkits (but the recognition of widgets is not so portable).
Plus Qt offers an easy way to integrate scripting into your app (if it is open source or you are developing it).
RE[3]: Two distinct beasts...
by benjymouse on Mon 8th Aug 2011 17:03
in reply to "RE[2]: Two distinct beasts..."
Actually, no it isn't... at least on Windows, it's quite easy:
http://www.autoitscript.com/site/autoit
The only exception to this is if apps don't use standard widgets, which is one of about 486472343 reasons why I'm a fan of standards in this regard.
http://www.autoitscript.com/site/autoit
The only exception to this is if apps don't use standard widgets, which is one of about 486472343 reasons why I'm a fan of standards in this regard.
Relying on widget types and -IDs for automation is inherently error prone and extremely brittle. I can never be more than a band-aid in the absence of a real automation framework.
All types of problems crop up: When suddenly an error message is displayed, when the dialog flow changes because of some unforeseen state, when timing issues crop up. Not to mention how enormously difficult it is to have robust error control in such an automated script. Or when a new version is released and IDs or widget types changes, when drop-down boxes doesn't contain the expected (and hard-wired) values. etc. etc.
Of course, on Linux, this would never work, since the general assumption seems to be that having 900 different toolkits is a good thing. So if GUI automation is hard, it's probably because of a piss-poor implementation, or someone hasn't written the tools to do it. But doesn't that work the same way on CLIs? I mean, SOMEONE has to build in pipes and stuff if you really want to do something useful with it ...
I wouldn't fault Linux for this. You were never supposed to rely on widget types and/or ID in the first place. The only ones who can reliably use those are the very developers of the apps who can use it for automated testing. The rest of us should stay away from it.
A much better solution is being developed with PowerShell. PowerShell "cmdlets" are suitable for sharing implementation with GUI application "command" patterns. Indeed this is what Exchange and several other MS server apps do now: They implement the cmdlets *first* and then build a GUI on top of those. The GUI then becomes the composition tool for PowerShell command pipelines.
The Exchange admin GUI will even show you the very PowerShell command pipeline it will execute and let use cut-and-paste it for automation.
PowerShell cmdlets are designed (unlike the typical *nix shell) for in-process execution, hence they can consume and produce strongly-typed in-process objects which makes it much more usable for actual GUI automation. In essense the GUI becomes a way to combine and design CLI commands.
RE[4]: Two distinct beasts...
by WorknMan on Tue 9th Aug 2011 02:42
in reply to "RE[3]: Two distinct beasts..."
All types of problems crop up: When suddenly an error message is displayed, when the dialog flow changes because of some unforeseen state, when timing issues crop up.
I've never had an issue with timing when using controlsend to a certain widget/id. But, as you state, you may have to make some minor modifications if/when a new version is released and the developer changes things that directly affects your script, but wouldn't that be the case for CLI apps as well, if they change the command-line syntax?
Of course, GUI automation isn't the ideal way to get things done; I was just pointing out that it's not exceedingly difficult either, and certainly not impossible. In fact, I do it all the time
Edited 2011-08-09 02:44 UTC





Member since:
2005-11-13
Well, GUI is not inherently a superset of CLI - it's a COMPANION. Meaning, each is better at some tasks than the other, so they go together like sugar and kool-aid, and I wouldn't be without either. Why must everything be turned into a pissing contest?
When it comes to GUIs, some have said how it's easier to pull up files in a CLI, like when you need to copy all files that are older than a certain date and with a certain extension. But what if you need to copy about 50 random files from a directory of 300 files that don't have any specific pattern to them? In that case, it's easy to CTRL+CLICK select the ones you need and drag them wherever. Similarly, the article mentions how it would be easier to manipulate music playlists, but what if you've got a playlist with 300 songs, and you want to custom sort them? Again, it's going to be easier to just drag stuff where you want them.
Of course, one could list dozens of examples were CLIs would be more useful... as I said before, it's not about which one is 'better', because that depends on the context.
Actually, no it isn't... at least on Windows, it's quite easy:
http://www.autoitscript.com/site/autoit
The only exception to this is if apps don't use standard widgets, which is one of about 486472343 reasons why I'm a fan of standards in this regard. Of course, on Linux, this would never work, since the general assumption seems to be that having 900 different toolkits is a good thing. So if GUI automation is hard, it's probably because of a piss-poor implementation, or someone hasn't written the tools to do it. But doesn't that work the same way on CLIs? I mean, SOMEONE has to build in pipes and stuff if you really want to do something useful with it ...