Undo in "Check Spelling" tool?

[kovidgoyal]

See my previous posts for why I dont think creating a checkpoint before starting spell check is particularly useful, but, since a number of people have asked for it, here you go:

https://github.com/kovidgoyal/calibr...4bdf80b6442ef9

[Phssthpok]

OK, thanks. However, this still means reverting to the start of the spellcheck and repeating maybe 100+ previous corrections in order to recover from a single error. A "last change undo" would be much more useful IMHO, as an individual spelling correction is basically a quick way of doing a "replace all" (which would be undoable).

[Phssthpok]

Originally Posted by kovidgoyal

Because nothing in life is free. Creating checkpoints after ever replace would add a huge performance cost.

But then why can it be done for "replace all" in the search dialog but not for the equivalent global replace in the spellcheck dialog? Or in other words, why is the spellcheck checkpoint a huge performance hit compared to the "replace all" checkpoint?

[kovidgoyal]

Think about the differences in the way that replace all and replace inside the spell check are used.

[eschwartz]

Kovid -- for the change.

I never thought that checkpoints for each spellcheck correction would be worth it.
I think, however, that this (one before starting spellcheck) will still come in useful, without the steep cost. Much appreciated.



@Phssthpok,
For one, there is no opportunity to do something similar before "a series of find & replaces". The only time the editor knows you are using one is when you actually run a specific S&R.
And S&R is far more flexible, and thus error-prone, than an extremely limited "replace with selected valid word".
So there was never a question of whether or not to checkpoint before global S&R.

[Phssthpok]

Originally Posted by kovidgoyal

Think about the differences in the way that replace all and replace inside the spell check are used.

Sorry, you need to give me more of a clue than that.

Spellcheck says "replace xxx by yyy" (to quote you).
It seems to me that this is just like using "replace all" to replace (\W)xxx(\W) by \1yyy\2, which can then be undone if necessary.

So why should it be fantastically more difficult to make the spellcheck change undoable?

Sorry to be a pain, but I really don't understand the problem here.

[kovidgoyal]

In spell check multiple replaces typically happen quickly, you can essentially click the replace button one after the other with no intervening steps, if the topmost suggestion is correct.

In replace all you need to enter a separate search and replace term each time.

That means that the cost of creating checkpoints is relatively much higher in the case of spell check, compared tot he typical time spent per replacement.

[theducks]

I dislike (full) Auto-anythings. (I don't need help messing up. I am very good at it, I have lots of practice )
I launch a SAVE before I do another bulk operation... (After checking what I do have )

Yes, I have had to retrieve a clean copy from the archive/device, on ocasion.

So thank Kovid for leaving the decision of when to save to me.

[Phssthpok]

Originally Posted by kovidgoyal

In spell check multiple replaces typically happen quickly, you can essentially click the replace button one after the other with no intervening steps, if the topmost suggestion is correct.

Mmmm. You can also do it all too easily when the topmost suggestion is NOT correct, as I've found to my cost...

How about making it an option, disabled by default, which clumsy idiots like me can turn on if we're prepared to accept the performance hit?

[Terisa de morgan]

Originally Posted by Phssthpok

How about making it an option, disabled by default, which clumsy idiots like me can turn on if we're prepared to accept the performance hit?

My two cents: You know about the performance. In a bit, the performance hit will be forgotten and users will complain about calibre being slower because of the checkpoint.

[Phssthpok]

Originally Posted by Terisa de morgan

My two cents: You know about the performance. In a bit, the performance hit will be forgotten and users will complain about calibre being slower because of the checkpoint.

They will only complain if they turn on the option deliberately (which should be accompanied by a tooltip warning like e.g. the "save metadata to file" option) and then forget they've done so.

I would just like the option (nothing compulsory here) not to have to redo two hours work every time my mouse button bounces, even if if costs me a couple of seconds on each change. As it is, I'm compelled to always "get it right" or to create manual checkpoints after each change (which takes longer to do than an automated checkpoint and can be forgotten).

But I'm starting to feel like a voice in the wilderness. Am I really the only clumsy one when it comes to running spell checks?

[kovidgoyal]

Here you go, without using checkpoints:

https://github.com/kovidgoyal/calibr...4477c2487b6b16

[BetterRed]

Originally Posted by kovidgoyal

Here you go, without using checkpoints:

https://github.com/kovidgoyal/calibr...4477c2487b6b16

Good compromise - an "Oops" button

BR

[Phssthpok]

Originally Posted by kovidgoyal

Here you go, without using checkpoints:

https://github.com/kovidgoyal/calibr...4477c2487b6b16

Excellent, you're my hero (again). Now a single slip won't be a total disaster.

[Under the Covers]

Just noticed this thread, as one who has had to revert to prior checkpoint and start again. Thanks for the "oops button" -- like that name.