File object enhancements #1180

skurfer · 2012-10-17T18:08:56Z

This brings together one of @tiennou's branches with one of mine.

He made a number of improvements to the way file objects are dealt with, including caching frequently used attributes to avoid repeated disk access. Both of us have removed a lot of unused code. I've also rebased it against the current master, so it shouldn't conflict.

My changes were mostly about allowing more control over preview icons. See #1177 for details. The one difference from that is I've restored the old default behavior, so previews will be generated for everything. Users have the option to control previews via a hidden preference. To test, try something like

defaults write com.blacktree.Quicksilver QSFilePreviewTypes '(public.image, public.movie, public.audio, com.adobe.pdf)'

This also includes all the commits from #1119 except for 16cb992, as it sounds like that was unresolved.

This allows any future operations that depend on UTI to grab it from memory instead of (potentially) hitting the disk again.

Other changes: * Unused variables/statements removed * Code specific to preference panes removed (It was never hit anyway) * Attempts to `initWithContentsOfFile:` removed (It was never getting hit - Quick Look takes care of it) * Special case code from plug-ins is now tried first (and based on UTI instead of extension) close #1149

- Apply DRY on those that do the same thing. The specific check for 'fold' was dropped because it's a standard type. - Replace some UTI constants.

I don't want to know what would happen in case the file wasn't on the same drive than ~/.Trash… Hopefully it looks it's only used by the update code when moving the old QS app to the Trash.

Conflicts: Quicksilver/Code-QuickStepCore/QSObject_FileHandling.h

Provide accessors for `-isPackage`, `-isAlias`, `-fileExtension`, and `-fileUTI`. Also changes `-isApplication` & `-isFolder` to use the cached LSItemInfoRecord.

You should never touch an NSError object before checking that an error was triggered, in case the callee decides to set the pointer to garbage.

This would force a restart in case a plugin declares one of those and gets loaded late.

This is a special case for pref panes that draws the pane's icon in the old-style preference pane template. Now that we have _real_ previewing for those, that's useless ;-).

Conflicts: Quicksilver/Code-QuickStepCore/QSObject_FileHandling.m

If we start assigning UTIs to all objects, this method could potentially be expanded to evaluate everything (and moved to the main QSObject code).

Users will have the ability to override this by setting a hidden default.

Developers that aren't on 10.8 will be missing this macro.

skurfer · 2012-10-17T18:38:19Z

There's another potential area for optimization I noticed while resolving conflicts and looking through the changes. -[QSDirectoryParser objectsFromPath:depth:types:excludeTypes:descend:] looks at a lot of the things that are now cached (but it does so prior to creating the QSObject from the file, so the new LSItemInfoRecord cache can't be used). I wonder if we could move things around a bit to avoid reading the same information twice.

tiennou · 2012-10-17T21:37:32Z

Be aware that 70d34ac already changes the method. I don't know if it's what NSWorkspace uses, and I still haven't checked how it handles trashing files from network drives. In fact, I was keen to have that checked before, in order to cut the amount of code churn, but that's not really a problem.

skurfer · 2012-10-18T12:38:48Z

Be aware that 70d34ac already changes the method.

Yeah, but I didn't think anyone was disputing that change.

I don't know if it's what NSWorkspace uses, and I still haven't checked how it handles trashing files from network drives.

It makes the sound, but nothing happens. So that answers that. I guess we need another way.

This reverts commit 170bfbc.

This reverts commit 9c4c4d0. Conflicts: Quicksilver/Code-QuickStepCore/QSObject_FileHandling.m

* only play the sound if at least one file was trashed successfully * if any files couldn't be trashed, display a notification listing them

skurfer · 2012-11-06T18:30:36Z

OK, I've updated the Trash action. Take a look.

skurfer · 2012-11-09T18:37:02Z

Some small issues with this:

The infoRecord method checks for a valid path before checking for cached results. This goes to disk on every cal, which is too expensive. I'm going to change it to check the cache first. If the file no longer exists, that should get picked up elsewhere.
To the greatest extent possible, we want all the disk reads necessary to populate the infoRecord dict to take place during catalog scanning and not in real-time as someone uses the app. The dict isn't preserved across relaunches, but the file objects themselves are, which means they aren't rescanned and the infoRecord isn't repopulated until the user does something with the object in the UI. I'd like to see if we can cache the dict to disk with the objects.

…le for a file

This can be a bit slower than the old way if a lot of file objects don't have infoRecord cached, but it's not terrible (about 0.1s), and once the info is cached, it's about 10 times faster than the previous code. This is a proof-of-concept. There are probably many other validation methods we could simplify.

Also, remove the now unused universalApps code. close #1184

skurfer · 2012-11-09T20:34:48Z

OK, the commit to add all applications to “Open With…” is in. Let me know if there’s anything else.

pjrobertson · 2012-11-10T13:02:32Z

There's something in here (or possible somewhere else between ß70 and here) that breaks right arrowing into text files.

I guess it's most likely here since this is where the fiddling with [QSObject loadChildrenForObject:] went on ;-)

pjrobertson · 2012-11-10T13:14:39Z

OK, here's the problem:
https://github.com/skurfer/Quicksilver/commit/c09deabb969b478582f0db7a3e98dcee55890d82

Caught you @tiennou ;-)

Seems like the return NO; shouldn't be there, otherwise the method never looks at the other handlers that could load children (like the text parser)

For @tiennou

And on the matter of that pull request: what if there's a method that say… returns 'NO' if there is an error, or if there's nothing to do or some other reason.

For example, take a look at: https://developer.apple.com/library/mac/#documentation/Cocoa/Reference/Foundation/Classes/NSFileManager_Class/Reference/Reference.html

This method returns NO if an error occurred OR "if a file, directory, or link already exists at path."
So in this case, how would you check to see if an error occurs? By checking if(error != nil) {} no?!

pjrobertson · 2012-11-10T13:19:26Z

Aaaand…. one more thing :)

I don't think that NSLog should be there at all. Try right arrowing into any file that shouldn't be right arrowed into (e.g. a .zip file or .jpg etc.) and watch your console.
Yucky yucky yuck!

Even if you right arrow into a text file, you still get that error log :(

pjrobertson · 2012-11-10T13:34:42Z

OK, last comment, I promise… :)

Just wrap lines 402 - 426 in an if ([object isFolder]) {} statement and we're all good :)

Should that method be isDirectory, since that's seems to be the convention, not isFolder?

skurfer · 2012-11-10T14:05:45Z

Just wrap lines 402 - 426 in an if ([object isFolder]) {} statement and we're all good :)

I can take care of that.

Should that method be isDirectory, since that's seems to be the convention, not isFolder?

For what it's worth, I had some weird problems when changing the validation around and it was because I was using isDirectory (thinking that was the name). There is a method by that name already, so we'd need need to be specific about the type of object in the loop, but I should have done that anyway.

So sounds like two votes for isDirectory. I'll fix that and the type assigned to the objects during validation.

pjrobertson · 2012-11-10T14:25:02Z

So sounds like two votes for isDirectory. I'll fix that and the type
assigned to the objects during validation

Awesome, cheers :)

On 10 November 2012 14:05, Rob McBroom notifications@github.com wrote:

Just wrap lines 402 - 426 in an if ([object isFolder]) {} statement and
we're all good :)

I can take care of that.

Should that method be isDirectory, since that's seems to be the
convention, not isFolder?

For what it's worth, I had some weird problems when changing the
validation around and it was because I was using isDirectory (thinking
that was the name). There is a method by that name already, so we'd need
need to be specific about the type of object in the loop, but I should have
done that anyway.

So sounds like two votes for isDirectory. I'll fix that and the type
assigned to the objects during validation.

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/1180#issuecomment-10255444.

skurfer · 2012-11-10T14:35:26Z

Two new commits.

pjrobertson · 2012-11-10T14:41:54Z

Cool. I'll run with it a little longer incase anything else crops up

tiennou · 2012-11-11T13:17:09Z

Sorry guys :-(. @skurfer your fix looks good.

Another quick thing : how do you feel about keeping -isFolder with the real "folder" meaning ? Something akin to return [self isDirectory] && !([self isPackage] || [self isApplication]. I'm sure that test is wrong but you see the point ;-).

pjrobertson · 2012-11-11T14:24:35Z

Another quick thing : how do you feel about keeping -isFolder with the real "folder" meaning ? Something akin to return [self isDirectory] && !([self isPackage] || [self isApplication]. I'm sure that test is wrong but you see the point ;-).

I think that's what isDirectory does now. Correct me if I'm wrong, but it
returns false if the file object is a package or application, yeah?

On 11 November 2012 13:17, Etienne Samson notifications@github.com wrote:

Sorry guys :-(. @skurfer https://github.com/skurfer your fix looks good.

Another quick thing : how do you feel about keeping -isFolder with the
real "folder" meaning ? Something akin to return [self isDirectory] &&
!([self isPackage] || [self isApplication]. I'm sure that test is wrong
but you see the point ;-).

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/1180#issuecomment-10266574.

skurfer · 2012-11-11T14:37:18Z

I think that's what isDirectory does now. Correct me if I'm wrong, but it returns false if the file object is a package or application, yeah?

Worth testing to find out. You can see I'm looking at isPackage too when populating the third pane for the Move/Copy To action.

tiennou · 2012-11-11T16:04:37Z

Yeah, the old tests look way too fishy for me. Testing using ß70, it looks to me .bundle things (a.k.a packages) get a folder quick icon, just before the preview kicks in and override the default. We might want to have both -isDirectory and -isFolder, in case we need precision : the former means low-level multi-children thing, the latter means the thing you expect to open in a new Finder window ;-). But IIRC (I reimplemented File Objects in Ruby, remember ;-)), applications would return YES to -isDirectory calls kind because of this...

skurfer · 2012-11-12T15:42:55Z

I've confirmed that isDirectory returnsYES for applications and other packages (which is what I expected). If you want, I can add an isFolder method. I think it just needs to do what the Move/Copy To actions are doing now, which is

return ([self isDirectory] && ![self isPackage]);

Applications are packages, so no need to test isApplication.

Looking for directories that aren't packages is a pretty common task.

pjrobertson · 2012-11-13T09:00:41Z

Fixed the conflict and pushed to master. Hopefully I got the merge conflict right ;-)

pjrobertson · 2012-11-13T09:00:54Z

Oh yeah - ß71 anyone? :)

skurfer · 2012-11-13T13:29:30Z

Oh yeah - ß71 anyone? :)

Heh. I thought this is all we were waiting on, but now there's some tempting stuff out there. :-)

Probably take me all day just to add this one pull to the change log.

skurfer and others added 23 commits Oct 15, 2012

don't blindly create a new instance of NSFileManager for every file

d016bd6

store the UTI with the object

0d7a299

This allows any future operations that depend on UTI to grab it from memory instead of (potentially) hitting the disk again.

DRY.

f3b6685

Group all UTI-related functions in QSUTI.h/m.

8026649

- Apply DRY on those that do the same thing. The specific check for 'fold' was dropped because it's a standard type. - Replace some UTI constants.

Use a thread to scan when clicking things in the interface.

5bfe8cc

Fix an issue with resolving an alias to a non-existent file.

9aff818

Use FSPathMoveObjectToTrashSync for moving files to the Trash.

70d34ac

I don't want to know what would happen in case the file wasn't on the same drive than ~/.Trash… Hopefully it looks it's only used by the update code when moving the old QS app to the Trash.

Catch exception when trying to instantiate instances.

1e777cb

Remove unused stuff + cleanup.

e108e70

Conflicts: Quicksilver/Code-QuickStepCore/QSObject_FileHandling.h

Use the object's cache to store its LSItemInfoRecord info.

a6794f5

Provide accessors for `-isPackage`, `-isAlias`, `-fileExtension`, and `-fileUTI`. Also changes `-isApplication` & `-isFolder` to use the cached LSItemInfoRecord.

Use the new accessors.

e17d52d

Fix usage of NSError.

c09deab

You should never touch an NSError object before checking that an error was triggered, in case the callee decides to set the pointer to garbage.

Remove static bundlePresetChildren cache.

30ac2da

This would force a restart in case a plugin declares one of those and gets loaded late.

Remove unused code.

49d8b81

This is a special case for pref panes that draws the pane's icon in the old-style preference pane template. Now that we have _real_ previewing for those, that's useless ;-).

Merge branch 'filesystem' into file-uti

52dd7aa

Conflicts: Quicksilver/Code-QuickStepCore/QSObject_FileHandling.m

use new method to get the UTI

1efc849

add a new method to see if a file object conforms to a UTI

9c4c4d0

If we start assigning UTIs to all objects, this method could potentially be expanded to evaluate everything (and moved to the main QSObject code).

use the new conformsToUTI: method when testing UTIs

170bfbc

remove more (unused) special code for Preference Panes

8286bce

load previews for everything by default (previous behavior)

d9a5bb9

Users will have the ability to override this by setting a hidden default.

Log the error loading message, and don't bother with the stack trace.

c037062

Use the AvailabilityMacros for this.

a938bd1

skurfer mentioned this pull request Oct 17, 2012

Fixes #1119

Closed

use the literal value for OS X 10.8 instead of the macro

b9fbc35

Developers that aren't on 10.8 will be missing this macro.

skurfer added 2 commits Oct 19, 2012

Revert "use the new conformsToUTI: method when testing UTIs"

a37c253

This reverts commit 170bfbc.

Revert "add a new method to see if a file object conforms to a UTI"

419a1d9

This reverts commit 9c4c4d0. Conflicts: Quicksilver/Code-QuickStepCore/QSObject_FileHandling.m

deal with failure in the Trash action

89a3d91

* only play the sound if at least one file was trashed successfully * if any files couldn't be trashed, display a notification listing them

skurfer added 4 commits Nov 9, 2012

check the cache in infoRecord first to avoid going to disk if possible

0d1672a

allow for the possibility that more than one icon provider is availab…

e2ea085

…le for a file

show all applications (in the catalog) for the Open With… action

4874f70

Also, remove the now unused universalApps code. close #1184

isFolder → isDirectory

03752f9

check that the object is a directory before treating it like one

6495cd9

add an isFolder method for file objects

bd5163a

Looking for directories that aren't packages is a pretty common task.

pjrobertson merged commit bd5163a into quicksilver:master Nov 13, 2012

pjrobertson mentioned this pull request Nov 13, 2012

Speedups #1206

Merged

skurfer mentioned this pull request Nov 28, 2012

File → Open With… refuses to show printers #1087

Closed

File object enhancements #1180

File object enhancements #1180

skurfer commented Oct 17, 2012

skurfer commented Oct 17, 2012

tiennou commented Oct 17, 2012

skurfer commented Oct 18, 2012

skurfer commented Nov 6, 2012

skurfer commented Nov 9, 2012

skurfer commented Nov 9, 2012

pjrobertson commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

skurfer commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

skurfer commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

tiennou commented Nov 11, 2012

pjrobertson commented Nov 11, 2012

skurfer commented Nov 11, 2012

tiennou commented Nov 11, 2012

skurfer commented Nov 12, 2012

pjrobertson commented Nov 13, 2012

pjrobertson commented Nov 13, 2012

skurfer commented Nov 13, 2012

File object enhancements #1180

File object enhancements #1180

Conversation

skurfer commented Oct 17, 2012

skurfer commented Oct 17, 2012

tiennou commented Oct 17, 2012

skurfer commented Oct 18, 2012

skurfer commented Nov 6, 2012

skurfer commented Nov 9, 2012

skurfer commented Nov 9, 2012

pjrobertson commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

skurfer commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

skurfer commented Nov 10, 2012

pjrobertson commented Nov 10, 2012

tiennou commented Nov 11, 2012

pjrobertson commented Nov 11, 2012

skurfer commented Nov 11, 2012

tiennou commented Nov 11, 2012

skurfer commented Nov 12, 2012

pjrobertson commented Nov 13, 2012

pjrobertson commented Nov 13, 2012

skurfer commented Nov 13, 2012