Very slow to compare significantly different directories

[vertigo220]

I have two directories full of subdirectories and more subdirectories, each full of pictures. I used another program to compare the two and remove all copies of images that are identical or near-identical (same image but rotated or tagged in one and not the other), which left me with ~15.5k files in one and 359 in the other. I then ran a comparison with WinMerge, which I expected to take a minute at most, since all it had to actually compare was those 359 files at most. However, it took several minutes, seeming to scan each and every file. Whatever method it uses for comparison doesn't seem to be very efficiently designed, and it seems there is a lot of room for improvement in speeding it up.

[sdottaka]

The default comparison method "full contents compare method" is not fast because it scans the file for detecting encoding and EOL type, even if only one of the files exists.

Try using "Binary contents" instead.

[vertigo220]

I considered that, but the help file says the full contents comparison is the most complete and recommended method, and just says "TBD" for binary compare, so I wasn't sure if it was a good idea. Two questions: what's the downside to using binary compare, and what purpose is there for checking all that in full compare if there's nothing to compare against. If there's no correlated file in the other folder(s), doing a "full" check on the one copy of the file that does exist seems pointless; it should just know it's "different" since it exists vs not exists.

[sdottaka]

what's the downside to using binary compare

"Ignore case" option and "Ignore spaces" option are not applied. It also does not apply line filters, comment filters, or plugins.

Below is a table showing the characteristics of each comparison method.

Method	Compare contents of file?	File size is used to determine the difference?	Modified date is used to determine the difference?	Detect the number of differences?	Binary file detection	Is "Ignore codepage differences" option applied?	Is "Ignore case" option is applied?	Is "Ignore spaces" option is applied	Is "Ignore carriage return differences" applied	Is "Filter comments" opiton is applied?	Line filters	Plugins	Speed
Full contens(default)	Yes	-	-	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Low
Quick contents	Yes	-	-	-	Yes	Yes	Yes	Yes	Yes	-	-	Yes	Low
Binary contents	Yes	Yes	-	-	-	-	-	-	-	-	-	-	Middle
Modifed Date	-	-	Yes	-	-	-	-	-	-	-	-	-	High
Modified Date and Size	-	Yes	Yes	-	-	-	-	-	-	-	-	-	High
Size	-	Yes	-	-	-	-	-	-	-	-	-	-	High

what purpose is there for checking all that in full compare if there's nothing to compare against.

Probably the purpose is to display EOL-type, etc. in the ListView column of the folder comparison window.

[vertigo220]

Probably the purpose is to display EOL-type, etc. in the ListView column of the folder comparison window.

So couldn't it just pass such files to make the comparison go much quicker then leave that line blank in the comparison, since it wouldn't matter anyways?

[sdottaka]

I think we could add an option to do so.
It would be nice if someone could Pull Request.

Hi, I have similar problem. But at least in my case, it's not about a speed of comparing files that are on both sides,
but that it takes long to go through files that are only on one side.

So far it looks to me, as if files that should be just quickly checked, that they are not on the other side,
are processed more.
Maybe read from disk, or something?

[sdottaka]

The "Full Contents" compare method, which is the default compare method, detects the encoding and judges whether the file is binary or not, even if one of the files does not exist, which causes the file to be read.
For faster folder comparison, use the "Binary Contents" compare method or the "Modifed Date and Size" compare method, which compares only the size and modified date of files.

So in that case, I want files that are on both sides to be compared with full content method,
and files that are only on one side to be compared with modified date and size method (or whichever doesn't read the file).

How do I set that up?

[sdottaka]

Unfortunately, there is no way to make such a comparison.

[vertigo220]

What about checking for a file to compare against before checking the encoding? IOW, currently it's doing checks on a file that take a while, then attempting to compare that file against another, whereas it would be much more efficient to see if there's even a file to compare against first before bothering with the other stuff, and only doing that if necessary, i.e. if there's a corresponding file.

[sdottaka]

The reason for not doing so is that in the Full Contents comparison method, we want the columns to show the encoding, EOL type, and whether the file is a binary file or not, even if the left and right files are not present.
However, if the columns for Encoding, EOL, etc. are hidden, it may be better to omit that process.

[vertigo220]

I suppose that could be a solution, but I see a couple issues with it. While I don't think I've ever used those columns, requiring the user to hide them to speed this up means two things: they have to re-show them any time they want to use them (and possibly waste time running a scan again if they ran it with the columns hidden then realized they wanted to see them) and it would mean it would run slow by default unless the user knew to hide those columns.

Personally, I can't imagine why one would want to see that info on a file when there's no corresponding file, since the whole point is to check differences, and if there's no file to compare to, the difference is the entire file. So it still seems it would be best to simply skip said files and leave those columns blank for them, with an option to alternate this behavior, in which case it would seem best to default it this way, and allow advanced users who actually have a need for this and therefore are willing to accept the performance hit to manually set it to the current behavior. Wouldn't this be possible?

Then again, maybe it would be best to simply use binary mode, and I'll have to give that a try with some test comparisons, but there are problems with that, mainly that it requires using not only a non-default, but also non-recommended mode to get good performance, which means most users aren't going to know to use it or won't know if they even should, just as I'm still unsure. Maybe I'm wrong, but it seems the default mode should provide the fastest performance while providing the results/info that most users will need, and if users need more info than that at a cost to performance, that should require changing options to their non-default settings, not to mention having such options to begin with.

I'll also admit that while I've used WinMerge quite a bit, there's likely a lot about it and its capabilities I don't know and don't have experience with, so it's possible I'm simply missing something that makes my arguments invalid. So if they make sense and you decide to change things accordingly, great, but if they don't, then so be it, in which case at least maybe there should be some sort of prompt on first-run explaining the performance and usage differences of full-compare vs binary and ask the user to choose the default method right then and there.