Account for existing backslashes in completion

The previous implementation assumed that there was no backslash escape
preceding the cursor when performing command line completion. This
caused a redundant backslash to be inserted when the next character in
the completed word needed escaping. After this commit, the shell checks
for an escape before the current position and removes it when inserting
the completion, leaving the correct number of backslashes.

Fixes #47

NEWS

@@ -5,6 +5,9 @@ History of Yash
   - [line-editing] Fixed the spurious error message printed when
     completing after `git config alias.` with the nounset shell option
     enabled.
+  - [line-editing] Completion no longer inserts a redundant backslash
+    to escape a character included in the completed word
+    when the cursor follows another backslash.
 
 ## Yash 2.56.1 (2024-03-20)
 

NEWS.ja

@@ -4,6 +4,9 @@ Yash 更新履歴
 
   - [行編集] nounset オプション有効時に `git config alias.` に続けて
     補完をしようとするとエラーが出るのを修正
+  - [行編集] カーソルがバックスラッシュの直後にある時に補完をすると
+    補完する単語に含まれるエスケープが必要な文字に対して
+    余計なバックスラッシュが挿入されるのを修正
 
 ## Yash 2.56.1 (2024-03-20)
 

lineedit/complete.c

@@ -1,6 +1,6 @@
 /* Yash: yet another shell */
 /* complete.c: command line completion */
-/* (C) 2007-2022 magicant */
+/* (C) 2007-2024 magicant */
 
 /* This program is free software: you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -486,10 +486,12 @@ void print_context_info(const le_context_T *ctxt)
 {
     const char *s DUMMY_INIT(NULL);
     switch (ctxt->quote) {
-        case QUOTE_NONE:    s = "none";    break;
-        case QUOTE_NORMAL:  s = "normal";  break;
-        case QUOTE_SINGLE:  s = "single";  break;
-        case QUOTE_DOUBLE:  s = "double";  break;
+        case QUOTE_NONE:            s = "none";            break;
+        case QUOTE_NORMAL:          s = "normal";          break;
+        case QUOTE_NORMAL_ESCAPED:  s = "normal escaped";  break;
+        case QUOTE_SINGLE:          s = "single";          break;
+        case QUOTE_DOUBLE:          s = "double";          break;
+        case QUOTE_DOUBLE_ESCAPED:  s = "double escaped";  break;
     }
     le_compdebug("quote type: %s", s);
     switch (ctxt->type & CTXT_MASK) {
@@ -1177,13 +1179,9 @@ size_t get_common_prefix_length(void)
  * up after this function. */
 void update_main_buffer(bool subst, bool finish)
 {
-    const le_candidate_T *cand;
-    xwcsbuf_T buf;
     size_t srclen;
     size_t substindex;
     le_quote_T quotetype;
-
-    wb_init(&buf);
     if (subst) {
         srclen = 0;
         substindex = ctxt->srcindex;
@@ -1192,8 +1190,29 @@ void update_main_buffer(bool subst, bool finish)
         srclen = wcslen(ctxt->src);
         substindex = ctxt->origindex;
         quotetype = ctxt->quote;
+
+        switch (quotetype) {
+            case QUOTE_NONE:
+            case QUOTE_NORMAL:
+            case QUOTE_SINGLE:
+            case QUOTE_DOUBLE:
+                break;
+            case QUOTE_NORMAL_ESCAPED:
+            case QUOTE_DOUBLE_ESCAPED:
+                // Get the backslash preceding the cursor removed
+                assert(substindex > 0);
+                substindex--;
+                assert(le_main_buffer.contents[substindex] == '\\');
+                break;
+        }
     }
+
+    // Quote the substring of the candidate to be inserted
+    const le_candidate_T *cand;
+    xwcsbuf_T quoted;
+    wb_init(&quoted);
     if (le_selected_candidate_index >= le_candidates.length) {
+        // No candidate selected. Quote the longest common prefix.
         size_t cpl = get_common_prefix_length();
         assert(srclen <= cpl);
         cand = le_candidates.contents[0];
@@ -1202,21 +1221,24 @@ void update_main_buffer(bool subst, bool finish)
         wchar_t value[valuelen + 1];
         wcsncpy(value, cand->origvalue + srclen, valuelen);
         value[valuelen] = L'\0';
-        quote(&buf, value, quotetype);
+        quote(&quoted, value, quotetype);
     } else {
+        // Quote the selected candidate.
         cand = le_candidates.contents[le_selected_candidate_index];
         assert(srclen <= wcslen(cand->origvalue));
         if (cand->origvalue[0] == L'\0' && quotetype == QUOTE_NORMAL)
-            wb_cat(&buf, L"\"\"");
+            wb_cat(&quoted, L"\"\"");
         else
-            quote(&buf, cand->origvalue + srclen, quotetype);
+            quote(&quoted, cand->origvalue + srclen, quotetype);
     }
+
+    // Insert the quoted candidate to the main buffer
     assert(le_main_index >= substindex);
     wb_replace_force(&le_main_buffer,
             substindex, le_main_index - substindex,
-            buf.contents, buf.length);
-    le_main_index = substindex + buf.length;
-    wb_destroy(&buf);
+            quoted.contents, quoted.length);
+    le_main_index = substindex + quoted.length;
+    wb_destroy(&quoted);
 
     if (le_selected_candidate_index >= le_candidates.length)
         return;
@@ -1227,11 +1249,13 @@ void update_main_buffer(bool subst, bool finish)
     switch (quotetype) {
         case QUOTE_NONE:
         case QUOTE_NORMAL:
+        case QUOTE_NORMAL_ESCAPED:
             break;
         case QUOTE_SINGLE:
             insert_to_main_buffer(L'\'');
             break;
         case QUOTE_DOUBLE:
+        case QUOTE_DOUBLE_ESCAPED:
             insert_to_main_buffer(L'"');
             break;
     }
@@ -1331,6 +1355,7 @@ void quote(xwcsbuf_T *restrict buf,
             wb_cat(buf, s);
             return;
         case QUOTE_NORMAL:
+        case QUOTE_NORMAL_ESCAPED:
             for (size_t i = 0; s[i] != L'\0'; i++) {
                 if (s[i] == L'\n') {
                     wb_ncat_force(buf, L"'\n'", 3);
@@ -1350,6 +1375,7 @@ void quote(xwcsbuf_T *restrict buf,
             }
             return;
         case QUOTE_DOUBLE:
+        case QUOTE_DOUBLE_ESCAPED:
             for (size_t i = 0; s[i] != L'\0'; i++) {
                 if (wcschr(CHARS_ESCAPABLE, s[i]) != NULL)
                     wb_wccat(buf, L'\\');

lineedit/complete.h

@@ -1,6 +1,6 @@
 /* Yash: yet another shell */
 /* complete.h: command line completion */
-/* (C) 2007-2022 magicant */
+/* (C) 2007-2024 magicant */
 
 /* This program is free software: you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -64,8 +64,10 @@ typedef struct le_candidate_T {
 typedef enum le_quote_T {
     QUOTE_NONE,    // no quotation required
     QUOTE_NORMAL,  // not in single/double quotation; backslashes can be used
+    QUOTE_NORMAL_ESCAPED,  // like QUOTE_NORMAL, but the cursor is backslashed
     QUOTE_SINGLE,  // in single quotation
     QUOTE_DOUBLE,  // in double quotation
+    QUOTE_DOUBLE_ESCAPED,  // like QUOTE_DOUBLE, but the cursor is backslashed
 } le_quote_T;
 typedef enum le_contexttype_T {
     CTXT_NORMAL,             // normal word

lineedit/compparse.c

@@ -1,6 +1,6 @@
 /* Yash: yet another shell */
 /* compparse.c: simple parser for command line completion */
-/* (C) 2007-2020 magicant */
+/* (C) 2007-2024 magicant */
 
 /* This program is free software: you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -618,7 +618,8 @@ wordunit_T *cparse_word(
             return NULL;
 
     wordunit_T *first = NULL, **lastp = &first;
-    bool indq = false;   /* in double quotes? */
+    bool indq = false;     /* in double quotes? */
+    bool escaped = false;  /* after a backslash? */
     size_t startindex = INDEX;
     size_t srcindex = le_main_index - (LEN - INDEX);
 
@@ -638,11 +639,14 @@ wordunit_T *cparse_word(
         case L'\0':
             goto done;
         case L'\\':
-            if (BUF[INDEX + 1] != L'\0') {
+            if (BUF[INDEX + 1] == L'\0') {
+                escaped = true;
+                INDEX += 1;
+                goto done;
+            } else {
                 INDEX += 2;
                 continue;
             }
-            break;
         case L'$':
             MAKE_WORDUNIT_STRING;
             wu = cparse_special_word_unit(
@@ -701,7 +705,9 @@ wordunit_T *cparse_word(
         assert(first != NULL);
         return first;
     } else {
-        pi->ctxt->quote = indq ? QUOTE_DOUBLE : QUOTE_NORMAL;
+        pi->ctxt->quote = indq ?
+            escaped ? QUOTE_DOUBLE_ESCAPED : QUOTE_DOUBLE :
+            escaped ? QUOTE_NORMAL_ESCAPED : QUOTE_NORMAL;
         goto finish;
     }