UI: Word Boundary Seeking #107762

Harley Acheson · 2023-05-08T22:37:25+02:00

Harley Acheson commented

2023-05-08 22:37:25 +02:00

Changes to how we detect boundaries between words when selecting,
moving the cursor, deleting, etc.

When moving any text cursor position to a word boundary we always first move in the intended direction by one character without any further testing. This results in unintended bad behavior. The following shows where your cursor is moved when you use ctrl-arrow keys to move to the next word boundary. You will see that the movements shown in red are inconsistent with all other movement:

This incorrect boundary movement causes errors when selecting words near the ends of word positions and if the word is only a single letter long. Following shows three types of word selections that accidentally include whitespace:

This happens because our "Cursor Step" functions take a use_init_step argument that forces this initial one-character movement. This argument was added so that movement between words would occur a word at a time. For example ctrl-left arrow will move from word start to previous word start, rather than word start, then word end, then word start.

However this movement should only happen when at whitespace. And we always call the "Cursor Step" functions with this set to true. So this PR just removes this argument entirely and does the initial movement only when appropriate.

Changes to how we detect boundaries between words when selecting, moving the cursor, deleting, etc. --- When moving any text cursor position to a word boundary we always first move in the intended direction by one character without any further testing. This results in unintended bad behavior. The following shows where your cursor is moved when you use ctrl-arrow keys to move to the next word boundary. You will see that the movements shown in red are inconsistent with all other movement: ![WordBoundary1.png](/attachments/547c8989-1a4e-47cb-8422-2ca84937a24e) This incorrect boundary movement causes errors when selecting words near the ends of word positions and if the word is only a single letter long. Following shows three types of word selections that accidentally include whitespace: ![WordBoundary2.gif](/attachments/af13cd1c-af00-446e-a400-f9e016cab3da) This happens because our "Cursor Step" functions take a `use_init_step` argument that forces this initial one-character movement. This argument was added so that movement between words would occur a word at a time. For example ctrl-left arrow will move from word start to previous word start, rather than word start, then word end, then word start. However this movement should only happen when at whitespace. And we always call the "Cursor Step" functions with this set to true. So this PR just removes this argument entirely and does the initial movement only when appropriate.

WordBoundary1.png

44 KiB

WordBoundary2.gif

46 KiB

Harley Acheson added 1 commit 2023-05-08 22:37:31 +02:00

f440dc1328 UI: Word Boundary Seeking

Changes to how we detect boundaries between words when selecting,
moving the cursor, deleting, etc.

Harley Acheson added this to the User Interface project 2023-05-08 22:37:44 +02:00

Harley Acheson requested review from Campbell Barton 2023-05-08 22:37:51 +02:00

Campbell Barton reviewed 2023-05-09 01:27:35 +02:00

source/blender/blenlib/intern/string_cursor_utf8.c Outdated

						
				@ -145,3 +144,3 @@

				  if (direction == STRCUR_DIR_NEXT) {

				    if (use_init_step) {

				    if (jump == STRCUR_JUMP_DELIM && ELEM(str[*pos], ' ', '\n', '\t')) {

Campbell Barton commented

2023-05-09 01:27:35 +02:00

Use cursor_delim_type_unicode(..) instead & check for STRCUR_DELIM_WHITESPACE.

Use `cursor_delim_type_unicode(..)` instead & check for `STRCUR_DELIM_WHITESPACE`.

Campbell Barton requested changes 2023-05-09 01:46:31 +02:00

Campbell Barton left a comment

This breaks selecting white-space for the text-editor, text-fields and 3d-text (not the Python console) try add multiple spaces then double-click to select the space between words. From what I cn see the white-space check should check adjacent character is not white-space.

Also, please leave changes to double-click word selection out of this patch, the checks for ELEM(..,' ', '\n', '\t') aren't great it's probably better to have a function as part of the BLI_str_cursor_* API, then there is no need to do character level checks & forwards/backwards motion. There is also less likelihood of behavior being inconsistent for text editing fields (currently the case with this patch, maybe a simple mistake/change).

This breaks selecting white-space for the text-editor, text-fields and 3d-text (not the Python console) try add multiple spaces then double-click to select the space between words. From what I cn see the white-space check should check adjacent character is not white-space. ---- Also, please leave changes to double-click word selection out of this patch, the checks for `ELEM(..,' ', '\n', '\t')` aren't great it's probably better to have a function as part of the `BLI_str_cursor_*` API, then there is no need to do character level checks & forwards/backwards motion. There is also less likelihood of behavior being inconsistent for text editing fields _(currently the case with this patch, maybe a simple mistake/change)._

Harley Acheson added 1 commit 2023-05-10 01:54:01 +02:00

a7fd8f1ffc Changes suggested by review.

Harley Acheson commented

2023-05-10 02:01:19 +02:00

@ideasman42 -... selecting white-space for the text-editor, text-fields and 3d-text (not the Python console)...

Yikes, it takes quite a bit more to get consistent behavior everywhere and deal with groups of multiple whitespaces correctly. Mostly that BLI_str_cursor_step_utf8 has to seek over multiple whitespace characters rather than that single character move.

Also, please leave changes to double-click word selection out of this patch.

Lots of current code, including the double-click for word selection rely on current behavior that needs to change. This does get rid of all those whitespace checks by instead using cursor_delim_type_unicode that I make public here as BLI_str_cursor_delim_type_utf8/32.

But yeah, lots of change here to fix something that is only slightly wonky now.

@ideasman42 -... selecting white-space for the text-editor, text-fields and 3d-text (not the Python console)... Yikes, it takes quite a bit more to get consistent behavior everywhere and deal with groups of multiple whitespaces correctly. Mostly that `BLI_str_cursor_step_utf8` has to seek over multiple whitespace characters rather than that single character move. > Also, please leave changes to double-click word selection out of this patch. Lots of current code, including the double-click for word selection rely on current behavior that needs to change. This does get rid of all those whitespace checks by instead using `cursor_delim_type_unicode` that I make public here as BLI_str_cursor_delim_type_utf8/32. But yeah, lots of change here to fix something that is only slightly wonky now.

Campbell Barton requested changes 2023-05-10 03:41:20 +02:00

source/blender/blenlib/BLI_string_cursor_utf8.h Outdated

						
				@ -11,6 +11,17 @@

				extern "C" {

				#endif

				typedef enum eStrCursorDelimType {

Campbell Barton commented

2023-05-10 03:40:40 +02:00

I'd rather avoid exposing internal implementation details and instead have a BLI_str_cursor_step_range(..) or similar.

Having to duplicate this logic about seems unnecessarily involved when it can be handled by a reusable utility function.

  txt_jump_right(text, true, use_init_step);
  if (!line->line[text->curc] ||
      (BLI_str_cursor_delim_type_utf8(line->line, line->len, text->curc) ==
       STRCUR_DELIM_WHITESPACE) ||
      (text->curc > 0 && BLI_str_cursor_delim_type_utf8(line->line, line->len, text->curc - 1) !=
                             STRCUR_DELIM_WHITESPACE))
  {
    /* Move left if to left is non-whitespace or if right is whitespace. */
    txt_jump_left(text, false);
  }
  txt_jump_right(text, true);

In fact, a BLI_str_cursor_step_range(..) utility function could be added as a separate refactor (it can use the current logic, which then simplifies this patch).

I'd rather avoid exposing internal implementation details and instead have a `BLI_str_cursor_step_range(..)` or similar. Having to duplicate this logic about seems unnecessarily involved when it can be handled by a reusable utility function. ``` txt_jump_right(text, true, use_init_step); if (!line->line[text->curc] || (BLI_str_cursor_delim_type_utf8(line->line, line->len, text->curc) == STRCUR_DELIM_WHITESPACE) || (text->curc > 0 && BLI_str_cursor_delim_type_utf8(line->line, line->len, text->curc - 1) != STRCUR_DELIM_WHITESPACE)) { /* Move left if to left is non-whitespace or if right is whitespace. */ txt_jump_left(text, false); } txt_jump_right(text, true); ``` ---- In fact, a `BLI_str_cursor_step_range(..)` utility function could be added as a separate refactor (it can use the current logic, which then simplifies this patch).

Harley Acheson commented

2023-05-10 18:28:39 +02:00

@ideasman42 - I'd rather avoid exposing internal implementation details and instead have...

Yes, it is much nicer like that. Added BLI_str_cursor_at_word_boundary_utfx() to clean that up and it gets rid of a lot of mess there.

@ideasman42 - I'd rather avoid exposing internal implementation details and instead have... Yes, it is much nicer like that. Added BLI_str_cursor_at_word_boundary_utfx() to clean that up and it gets rid of a lot of mess there.

Harley Acheson added 1 commit 2023-05-10 18:23:37 +02:00

e7ab62b8c4 Review changes

Campbell Barton commented

2023-05-12 02:42:50 +02:00

After some discussion this patch makes functional changes to cursor stepping (stepping over spaces works differently for e.g.) that should be made carefully and aren't directly related to resolving the issue selecting words.

Proposed solution is to add BLI_str_cursor_step_bounds_utf* which takes a point and returns a bounding range. This can be done without any changes to existing cursor stepping functions.

Internally it can step over matching character categories,

If pos && pos-1 categories match, seek in both directions, delimit by character category.
If they are different, use a priority check (where spaces happen to be lowest), and seek either forward or backwards in the direction with a higher priority: order could be something like [alpha-numeric, brackets, punctuation, space].

This way double clicking at a word boundary always prioritizes the words not the spaces.

Closing this PR as the proposed changes is quite a different direction from this one.

After some discussion this patch makes functional changes to cursor stepping (stepping over spaces works differently for e.g.) that should be made carefully and aren't directly related to resolving the issue selecting words. Proposed solution is to add `BLI_str_cursor_step_bounds_utf*` which takes a point and returns a bounding range. This can be done without any changes to existing cursor stepping functions. Internally it can step over matching character categories, - If `pos` && `pos-1` categories match, seek in both directions, delimit by character category. - If they are different, use a priority check (where spaces happen to be lowest), and seek either forward or backwards in the direction with a higher priority: order could be something like `[alpha-numeric, brackets, punctuation, space]`. This way double clicking at a word boundary always prioritizes the words not the spaces. ---- Closing this PR as the proposed changes is quite a different direction from this one.

Campbell Barton closed this pull request

2023-05-12 02:42:51 +02:00

Harley Acheson deleted branch WordBoundaries

2023-05-15 22:57:29 +02:00

Pull request closed

This pull request cannot be reopened because the branch was deleted.

Sign in to join this conversation.

No reviewers

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

UI: Word Boundary Seeking #107762

Pull request closed