Merge pull request #11833 from bfredl/set_text

nvim_buf_set_text

4 files changed, 474 insertions, 17 deletions

diff --git a/src/nvim/api/buffer.c b/src/nvim/api/buffer.c
index f1151d196a..8d82d22040 100644
--- a/src/nvim/api/buffer.c
+++ b/src/nvim/api/buffer.c
@@ -309,6 +309,27 @@ end:
   return rv;
 }
 
+static bool check_string_array(Array arr, bool disallow_nl, Error *err)
+{
+  for (size_t i = 0; i < arr.size; i++) {
+    if (arr.items[i].type != kObjectTypeString) {
+      api_set_error(err,
+                    kErrorTypeValidation,
+                    "All items in the replacement array must be strings");
+      return false;
+    }
+    // Disallow newlines in the middle of the line.
+    if (disallow_nl) {
+      const String l = arr.items[i].data.string;
+      if (memchr(l.data, NL, l.size)) {
+        api_set_error(err, kErrorTypeValidation,
+                      "String cannot contain newlines");
+        return false;
+      }
+    }
+  }
+  return true;
+}
 
 /// Sets (replaces) a line-range in the buffer.
 ///
@@ -362,22 +383,9 @@ void nvim_buf_set_lines(uint64_t channel_id,
     return;
   }
 
-  for (size_t i = 0; i < replacement.size; i++) {
-    if (replacement.items[i].type != kObjectTypeString) {
-      api_set_error(err,
-                    kErrorTypeValidation,
-                    "All items in the replacement array must be strings");
-      return;
-    }
-    // Disallow newlines in the middle of the line.
-    if (channel_id != VIML_INTERNAL_CALL) {
-      const String l = replacement.items[i].data.string;
-      if (memchr(l.data, NL, l.size)) {
-        api_set_error(err, kErrorTypeValidation,
-                      "String cannot contain newlines");
-        return;
-      }
-    }
+  bool disallow_nl = (channel_id != VIML_INTERNAL_CALL);
+  if (!check_string_array(replacement, disallow_nl, err)) {
+    return;
   }
 
   size_t new_len = replacement.size;
@@ -487,6 +495,250 @@ end:
   try_end(err);
 }
 
+/// Sets (replaces) a range in the buffer
+///
+/// This is recommended over nvim_buf_set_lines when only modifying parts of a
+/// line, as extmarks will be preserved on non-modified parts of the touched
+/// lines.
+///
+/// Indexing is zero-based and end-exclusive.
+///
+/// To insert text at a given index, set `start` and `end` ranges to the same
+/// index. To delete a range, set `replacement` to an array containing
+/// an empty string, or simply an empty array.
+///
+/// Prefer nvim_buf_set_lines when adding or deleting entire lines only.
+///
+/// @param channel_id
+/// @param buffer           Buffer handle, or 0 for current buffer
+/// @param start_row        First line index
+/// @param start_column     Last column
+/// @param end_row          Last line index
+/// @param end_column       Last column
+/// @param replacement      Array of lines to use as replacement
+/// @param[out] err         Error details, if any
+void nvim_buf_set_text(uint64_t channel_id, Buffer buffer,
+                       Integer start_row, Integer start_col,
+                       Integer end_row, Integer end_col,
+                       ArrayOf(String) replacement, Error *err)
+  FUNC_API_SINCE(7)
+{
+  FIXED_TEMP_ARRAY(scratch, 1);
+  if (replacement.size == 0) {
+    scratch.items[0] = STRING_OBJ(STATIC_CSTR_AS_STRING(""));
+    replacement = scratch;
+  }
+
+  buf_T *buf = find_buffer_by_handle(buffer, err);
+  if (!buf) {
+    return;
+  }
+
+  bool oob = false;
+
+  // check range is ordered and everything!
+  // start_row, end_row within buffer len (except add text past the end?)
+  start_row = normalize_index(buf, start_row, &oob);
+  if (oob || start_row == buf->b_ml.ml_line_count + 1) {
+    api_set_error(err, kErrorTypeValidation, "start_row out of bounds");
+    return;
+  }
+
+  end_row = normalize_index(buf, end_row, &oob);
+  if (oob || end_row == buf->b_ml.ml_line_count + 1) {
+    api_set_error(err, kErrorTypeValidation, "end_row out of bounds");
+    return;
+  }
+
+  char *str_at_start = (char *)ml_get_buf(buf, start_row, false);
+  if (start_col < 0 || (size_t)start_col > strlen(str_at_start)) {
+    api_set_error(err, kErrorTypeValidation, "start_col out of bounds");
+    return;
+  }
+
+  char *str_at_end = (char *)ml_get_buf(buf, end_row, false);
+  size_t len_at_end = strlen(str_at_end);
+  if (end_col < 0 || (size_t)end_col > len_at_end) {
+    api_set_error(err, kErrorTypeValidation, "end_col out of bounds");
+    return;
+  }
+
+  if (start_row > end_row || (end_row == start_row && start_col > end_col)) {
+    api_set_error(err, kErrorTypeValidation, "start is higher than end");
+    return;
+  }
+
+  bool disallow_nl = (channel_id != VIML_INTERNAL_CALL);
+  if (!check_string_array(replacement, disallow_nl, err)) {
+    return;
+  }
+
+  size_t new_len = replacement.size;
+
+  bcount_t new_byte = 0;
+  bcount_t old_byte = 0;
+
+  // calculate byte size of old region before it gets modified/deleted
+  if (start_row == end_row) {
+      old_byte = (bcount_t)end_col - start_col;
+  } else {
+      const char *bufline;
+      old_byte += (bcount_t)strlen(str_at_start) - start_col;
+      for (int64_t i = 1; i < end_row - start_row; i++) {
+          int64_t lnum = start_row + i;
+
+          bufline = (char *)ml_get_buf(buf, lnum, false);
+          old_byte += (bcount_t)(strlen(bufline))+1;
+      }
+      old_byte += (bcount_t)end_col+1;
+  }
+
+  String first_item = replacement.items[0].data.string;
+  String last_item = replacement.items[replacement.size-1].data.string;
+
+  size_t firstlen = (size_t)start_col+first_item.size;
+  size_t last_part_len = strlen(str_at_end) - (size_t)end_col;
+  if (replacement.size == 1) {
+    firstlen += last_part_len;
+  }
+  char *first = xmallocz(firstlen), *last = NULL;
+  memcpy(first, str_at_start, (size_t)start_col);
+  memcpy(first+start_col, first_item.data, first_item.size);
+  memchrsub(first+start_col, NUL, NL, first_item.size);
+  if (replacement.size == 1) {
+    memcpy(first+start_col+first_item.size, str_at_end+end_col, last_part_len);
+  } else {
+    last = xmallocz(last_item.size+last_part_len);
+    memcpy(last, last_item.data, last_item.size);
+    memchrsub(last, NUL, NL, last_item.size);
+    memcpy(last+last_item.size, str_at_end+end_col, last_part_len);
+  }
+
+  char **lines = (new_len != 0) ? xcalloc(new_len, sizeof(char *)) : NULL;
+  lines[0] = first;
+  new_byte += (bcount_t)(first_item.size);
+  for (size_t i = 1; i < new_len-1; i++) {
+    const String l = replacement.items[i].data.string;
+
+    // Fill lines[i] with l's contents. Convert NULs to newlines as required by
+    // NL-used-for-NUL.
+    lines[i] = xmemdupz(l.data, l.size);
+    memchrsub(lines[i], NUL, NL, l.size);
+    new_byte += (bcount_t)(l.size)+1;
+  }
+  if (replacement.size > 1) {
+    lines[replacement.size-1] = last;
+    new_byte += (bcount_t)(last_item.size)+1;
+  }
+
+  try_start();
+  aco_save_T aco;
+  aucmd_prepbuf(&aco, (buf_T *)buf);
+
+  if (!MODIFIABLE(buf)) {
+    api_set_error(err, kErrorTypeException, "Buffer is not 'modifiable'");
+    goto end;
+  }
+
+  // Small note about undo states: unlike set_lines, we want to save the
+  // undo state of one past the end_row, since end_row is inclusive.
+  if (u_save((linenr_T)start_row - 1, (linenr_T)end_row + 1) == FAIL) {
+    api_set_error(err, kErrorTypeException, "Failed to save undo information");
+    goto end;
+  }
+
+  ptrdiff_t extra = 0;  // lines added to text, can be negative
+  size_t old_len = (size_t)(end_row-start_row+1);
+
+  // If the size of the range is reducing (ie, new_len < old_len) we
+  // need to delete some old_len. We do this at the start, by
+  // repeatedly deleting line "start".
+  size_t to_delete = (new_len < old_len) ? (size_t)(old_len - new_len) : 0;
+  for (size_t i = 0; i < to_delete; i++) {
+    if (ml_delete((linenr_T)start_row, false) == FAIL) {
+      api_set_error(err, kErrorTypeException, "Failed to delete line");
+      goto end;
+    }
+  }
+
+  if (to_delete > 0) {
+    extra -= (ptrdiff_t)to_delete;
+  }
+
+  // For as long as possible, replace the existing old_len with the
+  // new old_len. This is a more efficient operation, as it requires
+  // less memory allocation and freeing.
+  size_t to_replace = old_len < new_len ? old_len : new_len;
+  for (size_t i = 0; i < to_replace; i++) {
+    int64_t lnum = start_row + (int64_t)i;
+
+    if (lnum >= MAXLNUM) {
+      api_set_error(err, kErrorTypeValidation, "Index value is too high");
+      goto end;
+    }
+
+    if (ml_replace((linenr_T)lnum, (char_u *)lines[i], false) == FAIL) {
+      api_set_error(err, kErrorTypeException, "Failed to replace line");
+      goto end;
+    }
+    // Mark lines that haven't been passed to the buffer as they need
+    // to be freed later
+    lines[i] = NULL;
+  }
+
+  // Now we may need to insert the remaining new old_len
+  for (size_t i = to_replace; i < new_len; i++) {
+    int64_t lnum = start_row + (int64_t)i - 1;
+
+    if (lnum >= MAXLNUM) {
+      api_set_error(err, kErrorTypeValidation, "Index value is too high");
+      goto end;
+    }
+
+    if (ml_append((linenr_T)lnum, (char_u *)lines[i], 0, false) == FAIL) {
+      api_set_error(err, kErrorTypeException, "Failed to insert line");
+      goto end;
+    }
+
+    // Same as with replacing, but we also need to free lines
+    xfree(lines[i]);
+    lines[i] = NULL;
+    extra++;
+  }
+
+  // Adjust marks. Invalidate any which lie in the
+  // changed range, and move any in the remainder of the buffer.
+  mark_adjust((linenr_T)start_row,
+              (linenr_T)end_row,
+              MAXLNUM,
+              (long)extra,
+              kExtmarkNOOP);
+
+  colnr_T col_extent = (colnr_T)(end_col
+                                 - ((end_row == start_row) ? start_col : 0));
+  extmark_splice(buf, (int)start_row-1, (colnr_T)start_col,
+                 (int)(end_row-start_row), col_extent, old_byte,
+                 (int)new_len-1, (colnr_T)last_item.size, new_byte,
+                 kExtmarkUndo);
+
+
+  changed_lines((linenr_T)start_row, 0, (linenr_T)end_row, (long)extra, true);
+
+  // adjust cursor like an extmark ( i e it was inside last_part_len)
+  if (curwin->w_cursor.lnum == end_row && curwin->w_cursor.col > end_col) {
+    curwin->w_cursor.col -= col_extent - (colnr_T)last_item.size;
+  }
+  fix_cursor((linenr_T)start_row, (linenr_T)end_row, (linenr_T)extra);
+
+end:
+  for (size_t i = 0; i < new_len; i++) {
+    xfree(lines[i]);
+  }
+  xfree(lines);
+  aucmd_restbuf(&aco);
+  try_end(err);
+}
+
 /// Returns the byte offset of a line (0-indexed). |api-indexing|
 ///
 /// Line 1 (index=0) has offset 0. UTF-8 bytes are counted. EOL is one byte.
diff --git a/src/nvim/extmark.c b/src/nvim/extmark.c
index ba685b158e..b2d8532cd7 100644
--- a/src/nvim/extmark.c
+++ b/src/nvim/extmark.c
@@ -560,6 +560,23 @@ void extmark_adjust(buf_T *buf,
                       new_row, 0, new_byte, undo);
 }
 
+// Adjust extmarks following a text edit.
+//
+// @param buf
+// @param start_row   Start row of the region to be changed
+// @param start_col   Start col of the region to be changed
+// @param old_row     End row of the region to be changed.
+//                      Encoded as an offset to start_row.
+// @param old_col     End col of the region to be changed. Encodes
+//                      an offset from start_col if old_row = 0; otherwise,
+//                      encodes the end column of the old region.
+// @param old_byte    Byte extent of the region to be changed.
+// @param new_row     Row offset of the new region.
+// @param new_col     Col offset of the new region. Encodes an offset from
+//                      start_col if new_row = 0; otherwise, encodes
+//                      the end column of the new region.
+// @param new_byte    Byte extent of the new region.
+// @param undo
 void extmark_splice(buf_T *buf,
                     int start_row, colnr_T start_col,
                     int old_row, colnr_T old_col, bcount_t old_byte,
diff --git a/test/functional/api/buffer_spec.lua b/test/functional/api/buffer_spec.lua
index 8ed642b43e..fb8ed6a9d7 100644
--- a/test/functional/api/buffer_spec.lua
+++ b/test/functional/api/buffer_spec.lua
@@ -392,6 +392,129 @@ describe('api/buf', function()
     end)
   end)
 
+  describe('nvim_buf_get_lines, nvim_buf_set_text', function()
+    local get_lines, set_text = curbufmeths.get_lines, curbufmeths.set_text
+
+    it('works', function()
+      insert([[
+      hello foo!
+      text
+      ]])
+
+      eq({'hello foo!'}, get_lines(0, 1, true))
+
+
+      -- can replace a single word
+      set_text(0, 6, 0, 9, {'world'})
+      eq({'hello world!', 'text'}, get_lines(0, 2, true))
+
+      -- can insert text
+      set_text(0, 0, 0, 0, {'well '})
+      eq({'well hello world!', 'text'}, get_lines(0, 2, true))
+
+      -- can delete text
+      set_text(0, 0, 0, 5, {''})
+      eq({'hello world!', 'text'}, get_lines(0, 2, true))
+
+      -- can replace with multiple lines
+      set_text(0, 6, 0, 11, {'foo', 'wo', 'more'})
+      eq({'hello foo', 'wo', 'more!', 'text'}, get_lines(0,  4, true))
+
+      -- will join multiple lines if needed
+      set_text(0, 6, 3, 4, {'bar'})
+      eq({'hello bar'}, get_lines(0,  1, true))
+    end)
+
+    it('works with undo', function()
+        insert([[
+        hello world!
+        foo bar
+        ]])
+
+        -- setting text
+        set_text(0, 0, 0, 0, {'well '})
+        feed('u')
+        eq({'hello world!'}, get_lines(0, 1, true))
+
+        -- deleting text
+        set_text(0, 0, 0, 6, {''})
+        feed('u')
+        eq({'hello world!'}, get_lines(0, 1, true))
+
+        -- inserting newlines
+        set_text(0, 0, 0, 0, {'hello', 'mr '})
+        feed('u')
+        eq({'hello world!'}, get_lines(0, 1, true))
+
+        -- deleting newlines
+        set_text(0, 0, 1, 4, {'hello'})
+        feed('u')
+        eq({'hello world!'}, get_lines(0, 1, true))
+    end)
+
+    it('updates the cursor position', function()
+      insert([[
+      hello world!
+      ]])
+
+      -- position the cursor on `!`
+      curwin('set_cursor', {1, 11})
+      -- replace 'world' with 'foo'
+      set_text(0, 6, 0, 11, {'foo'})
+      eq('hello foo!', curbuf_depr('get_line', 0))
+      -- cursor should be moved left by two columns (replacement is shorter by 2 chars)
+      eq({1, 9}, curwin('get_cursor'))
+    end)
+
+    it('can handle NULs', function()
+      set_text(0, 0, 0, 0, {'ab\0cd'})
+      eq('ab\0cd', curbuf_depr('get_line', 0))
+    end)
+
+    it('adjusts extmarks', function()
+      local ns = request('nvim_create_namespace', "my-fancy-plugin")
+      insert([[
+      foo bar
+      baz
+      ]])
+      local id1 = curbufmeths.set_extmark(ns, 0, 1, {})
+      local id2 = curbufmeths.set_extmark(ns, 0, 7, {})
+      local id3 = curbufmeths.set_extmark(ns, 1, 1, {})
+      set_text(0, 4, 0, 7, {"q"})
+
+      eq({'foo q', 'baz'}, get_lines(0, 2, true))
+      -- mark before replacement point is unaffected
+      eq({0, 1}, curbufmeths.get_extmark_by_id(ns, id1, {}))
+      -- mark gets shifted back because the replacement was shorter
+      eq({0, 5}, curbufmeths.get_extmark_by_id(ns, id2, {}))
+      -- mark on the next line is unaffected
+      eq({1, 1}, curbufmeths.get_extmark_by_id(ns, id3, {}))
+
+      -- replacing the text spanning two lines will adjust the mark on the next line
+      set_text(0, 3, 1, 3, {"qux"})
+      eq({'fooqux', ''}, get_lines(0, 2, true))
+      eq({0, 6}, curbufmeths.get_extmark_by_id(ns, id3, {}))
+      -- but mark before replacement point is still unaffected
+      eq({0, 1}, curbufmeths.get_extmark_by_id(ns, id1, {}))
+      -- and the mark in the middle was shifted to the end of the insertion
+      eq({0, 6}, curbufmeths.get_extmark_by_id(ns, id2, {}))
+
+      -- marks should be put back into the same place after undoing
+      set_text(0, 0, 0, 2, {''})
+      feed('u')
+      eq({0, 1}, curbufmeths.get_extmark_by_id(ns, id1, {}))
+      eq({0, 6}, curbufmeths.get_extmark_by_id(ns, id2, {}))
+      eq({0, 6}, curbufmeths.get_extmark_by_id(ns, id3, {}))
+
+	  -- marks should be shifted over by the correct number of bytes for multibyte
+	  -- chars
+	  set_text(0, 0, 0, 0, {'Ø'})
+	  eq({0, 3}, curbufmeths.get_extmark_by_id(ns, id1, {}))
+	  eq({0, 8}, curbufmeths.get_extmark_by_id(ns, id2, {}))
+	  eq({0, 8}, curbufmeths.get_extmark_by_id(ns, id3, {}))
+    end)
+  end)
+
   describe('nvim_buf_get_offset', function()
     local get_offset = curbufmeths.get_offset
     it('works', function()
diff --git a/test/functional/lua/buffer_updates_spec.lua b/test/functional/lua/buffer_updates_spec.lua
index 9f2b1b6e52..67dc5f5a16 100644
--- a/test/functional/lua/buffer_updates_spec.lua
+++ b/test/functional/lua/buffer_updates_spec.lua
@@ -288,7 +288,8 @@ describe('lua: nvim_buf_attach on_bytes', function()
     -- TODO: while we are brewing the real strong coffe,
     -- verify should check buf_get_offset after every check_events
     if verify then
-      meths.buf_get_offset(0, meths.buf_line_count(0))
+      local len = meths.buf_get_offset(0, meths.buf_line_count(0))
+      eq(len == -1 and 1 or len, string.len(shadowbytes))
     end
     exec_lua("return test_register(...)", 0, "test1", false, false, true)
     meths.buf_get_changedtick(0)
@@ -510,6 +511,70 @@ describe('lua: nvim_buf_attach on_bytes', function()
         { "test1", "bytes", 1, 3, 0, 1, 1, 0, 3, 3, 0, 1, 1 };
       }
     end)
+
+    it('nvim_buf_set_text insert', function()
+      local check_events = setup_eventcheck(verify, {"bastext"})
+      meths.buf_set_text(0, 0, 3, 0, 3, {"fiol","kontra"})
+      check_events {
+        { "test1", "bytes", 1, 3, 0, 3, 3, 0, 0, 0, 1, 6, 11 };
+      }
+
+      meths.buf_set_text(0, 1, 6, 1, 6, {"punkt","syntgitarr","övnings"})
+      check_events {
+        { "test1", "bytes", 1, 4, 1, 6, 14, 0, 0, 0, 2, 8, 25 };
+      }
+
+      eq({ "basfiol", "kontrapunkt", "syntgitarr", "övningstext" },
+         meths.buf_get_lines(0, 0, -1, true))
+    end)
+
+    it('nvim_buf_set_text replace', function()
+      local check_events = setup_eventcheck(verify, origlines)
+
+      meths.buf_set_text(0, 2, 3, 2, 8, {"very text"})
+      check_events {
+        { "test1", "bytes", 1, 3, 2, 3, 35, 0, 5, 5, 0, 9, 9 };
+      }
+
+      meths.buf_set_text(0, 3, 5, 3, 7, {" splitty","line "})
+      check_events {
+        { "test1", "bytes", 1, 4, 3, 5, 57, 0, 2, 2, 1, 5, 14 };
+      }
+
+      meths.buf_set_text(0, 0, 8, 1, 2, {"JOINY"})
+      check_events {
+        { "test1", "bytes", 1, 5, 0, 8, 8, 1, 2, 10, 0, 5, 5 };
+      }
+
+      meths.buf_set_text(0, 4, 0, 6, 0, {"was 5,6",""})
+      check_events {
+        { "test1", "bytes", 1, 6, 4, 0, 75, 2, 0, 32, 1, 0, 8 };
+      }
+
+      eq({ "originalJOINYiginal line 2", "orivery text line 3", "origi splitty",
+           "line l line 4", "was 5,6", "    indented line" },
+         meths.buf_get_lines(0, 0, -1, true))
+
+    end)
+
+    it('nvim_buf_set_text delete', function()
+      local check_events = setup_eventcheck(verify, origlines)
+
+      -- really {""} but accepts {} as a shorthand
+      meths.buf_set_text(0, 0, 0, 1, 0, {})
+      check_events {
+        { "test1", "bytes", 1, 3, 0, 0, 0, 1, 0, 16, 0, 0, 0 };
+      }
+
+      -- TODO(bfredl): this works but is not as convenient as set_lines
+      meths.buf_set_text(0, 4, 15, 5, 17, {""})
+      check_events {
+        { "test1", "bytes", 1, 4, 4, 15, 79, 1, 17, 18, 0, 0, 0 };
+      }
+      eq({ "original line 2", "original line 3", "original line 4",
+           "original line 5", "original line 6" },
+         meths.buf_get_lines(0, 0, -1, true))
+    end)
   end
 
   describe('(with verify) handles', function()