[readline] do a fuzzy match if the prefix match fails to find anything

Prototype for #541

NEWS

@@ -1,4 +1,14 @@
 
+lnav v0.8.5:
+     Interface Changes:
+     * The auto-complete behavior in the prompt has been modified to fall back
+       to a fuzzy search if the prefix search finds no matches.  For example,
+       typing in ":fin" and pressing TAB would previously not do anything.
+       Now, the ":fin" will be completed to ":filter-in " since that is a
+       strong fuzzy match.  If there are multiple matches, as would happen
+       with ":dfil", readline's menu-complete behavior will be engaged and
+       you can press TAB cycle through the options.
+
 lnav v0.8.4:
      Features:
      * Added the ':comment' command that can be used to attach a comment to a

src/CMakeLists.txt

@@ -13,6 +13,7 @@ set(diag_STAT_SRCS
     file_vtab.cc
     fs-extension-functions.cc
     fstat_vtab.cc
+    fts_fuzzy_match.cc
     grep_proc.cc
     help_text_formatter.cc
     hist_source.cc
@@ -108,6 +109,7 @@ set(diag_STAT_SRCS
     filter_observer.hh
     format-text-files.hh
     fstat_vtab.hh
+    fts_fuzzy_match.hh
     grep_highlighter.hh
     help.hh
     help_text_formatter.hh

src/Makefile.am

@@ -162,6 +162,7 @@ noinst_HEADERS = \
 	filter_observer.hh \
 	format-text-files.hh \
 	fstat_vtab.hh \
+	fts_fuzzy_match.hh \
 	grep_highlighter.hh \
 	grep_proc.hh \
 	help.hh \
@@ -279,6 +280,7 @@ libdiag_a_SOURCES = \
 	file_vtab.cc \
 	fstat_vtab.cc \
     fs-extension-functions.cc \
+    fts_fuzzy_match.cc \
 	grep_proc.cc \
 	help_text_formatter.cc \
 	hist_source.cc \

src/fts_fuzzy_match.cc

@@ -0,0 +1,178 @@
+// LICENSE
+//
+//   This software is dual-licensed to the public domain and under the following
+//   license: you are granted a perpetual, irrevocable license to copy, modify,
+//   publish, and distribute this file as you see fit.
+
+#include "config.h"
+
+#include "fts_fuzzy_match.hh"
+
+namespace fts {
+
+    // Forward declarations for "private" implementation
+    namespace fuzzy_internal {
+        static bool fuzzy_match_recursive(const char * pattern, const char * str, int & outScore, const char * strBegin,
+                                          uint8_t const * srcMatches,  uint8_t * newMatches,  int maxMatches, int nextMatch,
+                                          int & recursionCount, int recursionLimit);
+    }
+
+    // Public interface
+    bool fuzzy_match_simple(char const * pattern, char const * str) {
+        while (*pattern != '\0' && *str != '\0')  {
+            if (tolower(*pattern) == tolower(*str))
+                ++pattern;
+            ++str;
+        }
+
+        return *pattern == '\0' ? true : false;
+    }
+
+    bool fuzzy_match(char const * pattern, char const * str, int & outScore) {
+        uint8_t matches[256];
+        return fuzzy_match(pattern, str, outScore, matches, sizeof(matches));
+    }
+
+    bool fuzzy_match(char const * pattern, char const * str, int & outScore, uint8_t * matches, int maxMatches) {
+        int recursionCount = 0;
+        int recursionLimit = 10;
+
+        return fuzzy_internal::fuzzy_match_recursive(pattern, str, outScore, str, nullptr, matches, maxMatches, 0, recursionCount, recursionLimit);
+    }
+
+    // Private implementation
+    static bool fuzzy_internal::fuzzy_match_recursive(const char * pattern, const char * str, int & outScore,
+                                                      const char * strBegin, uint8_t const * srcMatches, uint8_t * matches, int maxMatches,
+                                                      int nextMatch, int & recursionCount, int recursionLimit)
+    {
+        // Count recursions
+        ++recursionCount;
+        if (recursionCount >= recursionLimit)
+            return false;
+
+        // Detect end of strings
+        if (*pattern == '\0' || *str == '\0')
+            return false;
+
+        // Recursion params
+        bool recursiveMatch = false;
+        uint8_t bestRecursiveMatches[256];
+        int bestRecursiveScore = 0;
+
+        // Loop through pattern and str looking for a match
+        bool first_match = true;
+        while (*pattern != '\0' && *str != '\0') {
+
+            // Found match
+            if (tolower(*pattern) == tolower(*str)) {
+
+                // Supplied matches buffer was too short
+                if (nextMatch >= maxMatches)
+                    return false;
+
+                // "Copy-on-Write" srcMatches into matches
+                if (first_match && srcMatches) {
+                    memcpy(matches, srcMatches, nextMatch);
+                    first_match = false;
+                }
+
+                // Recursive call that "skips" this match
+                uint8_t recursiveMatches[256];
+                int recursiveScore;
+                if (fuzzy_match_recursive(pattern, str + 1, recursiveScore, strBegin, matches, recursiveMatches, sizeof(recursiveMatches), nextMatch, recursionCount, recursionLimit)) {
+
+                    // Pick best recursive score
+                    if (!recursiveMatch || recursiveScore > bestRecursiveScore) {
+                        memcpy(bestRecursiveMatches, recursiveMatches, 256);
+                        bestRecursiveScore = recursiveScore;
+                    }
+                    recursiveMatch = true;
+                }
+
+                // Advance
+                matches[nextMatch++] = (uint8_t)(str - strBegin);
+                ++pattern;
+            }
+            ++str;
+        }
+
+        // Determine if full pattern was matched
+        bool matched = *pattern == '\0' ? true : false;
+
+        // Calculate score
+        if (matched) {
+            const int sequential_bonus = 15;            // bonus for adjacent matches
+            const int separator_bonus = 30;             // bonus if match occurs after a separator
+            const int camel_bonus = 30;                 // bonus if match is uppercase and prev is lower
+            const int first_letter_bonus = 15;          // bonus if the first letter is matched
+
+            const int leading_letter_penalty = -5;      // penalty applied for every letter in str before the first match
+            const int max_leading_letter_penalty = -15; // maximum penalty for leading letters
+            const int unmatched_letter_penalty = -1;    // penalty for every letter that doesn't matter
+
+            // Iterate str to end
+            while (*str != '\0')
+                ++str;
+
+            // Initialize score
+            outScore = 100;
+
+            // Apply leading letter penalty
+            int penalty = leading_letter_penalty * matches[0];
+            if (penalty < max_leading_letter_penalty)
+                penalty = max_leading_letter_penalty;
+            outScore += penalty;
+
+            // Apply unmatched penalty
+            int unmatched = (int)(str - strBegin) - nextMatch;
+            outScore += unmatched_letter_penalty * unmatched;
+
+            // Apply ordering bonuses
+            for (int i = 0; i < nextMatch; ++i) {
+                uint8_t currIdx = matches[i];
+
+                if (i > 0) {
+                    uint8_t prevIdx = matches[i - 1];
+
+                    // Sequential
+                    if (currIdx == (prevIdx + 1))
+                        outScore += sequential_bonus;
+                }
+
+                // Check for bonuses based on neighbor character value
+                if (currIdx > 0) {
+                    // Camel case
+                    char neighbor = strBegin[currIdx - 1];
+                    char curr = strBegin[currIdx];
+                    if (::islower(neighbor) && ::isupper(curr))
+                        outScore += camel_bonus;
+
+                    // Separator
+                    bool neighborSeparator = neighbor == '_' || neighbor == ' ';
+                    if (neighborSeparator)
+                        outScore += separator_bonus;
+                }
+                else {
+                    // First letter
+                    outScore += first_letter_bonus;
+                }
+            }
+        }
+
+        // Return best result
+        if (recursiveMatch && (!matched || bestRecursiveScore > outScore)) {
+            // Recursive score is better than "this"
+            memcpy(matches, bestRecursiveMatches, maxMatches);
+            outScore = bestRecursiveScore;
+            return true;
+        }
+        else if (matched) {
+            // "this" score is better than recursive
+            return true;
+        }
+        else {
+            // no match
+            return false;
+        }
+    }
+} // namespace fts

src/fts_fuzzy_match.hh

@@ -0,0 +1,47 @@
+// LICENSE
+//
+//   This software is dual-licensed to the public domain and under the following
+//   license: you are granted a perpetual, irrevocable license to copy, modify,
+//   publish, and distribute this file as you see fit.
+//
+// VERSION
+//   0.2.0  (2017-02-18)  Scored matches perform exhaustive search for best score
+//   0.1.0  (2016-03-28)  Initial release
+//
+// AUTHOR
+//   Forrest Smith
+//
+// NOTES
+//   Compiling
+//     You MUST add '#define FTS_FUZZY_MATCH_IMPLEMENTATION' before including this header in ONE source file to create implementation.
+//
+//   fuzzy_match_simple(...)
+//     Returns true if each character in pattern is found sequentially within str
+//
+//   fuzzy_match(...)
+//     Returns true if pattern is found AND calculates a score.
+//     Performs exhaustive search via recursion to find all possible matches and match with highest score.
+//     Scores values have no intrinsic meaning. Possible score range is not normalized and varies with pattern.
+//     Recursion is limited internally (default=10) to prevent degenerate cases (pattern="aaaaaa" str="aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa")
+//     Uses uint8_t for match indices. Therefore patterns are limited to 256 characters.
+//     Score system should be tuned for YOUR use case. Words, sentences, file names, or method names all prefer different tuning.
+
+
+#ifndef FTS_FUZZY_MATCH_H
+#define FTS_FUZZY_MATCH_H
+
+
+#include <cstdint> // uint8_t
+#include <ctype.h> // ::tolower, ::toupper
+#include <cstring> // memcpy
+
+#include <cstdio>
+
+// Public interface
+namespace fts {
+    bool fuzzy_match_simple(char const * pattern, char const * str);
+    bool fuzzy_match(char const * pattern, char const * str, int & outScore);
+    bool fuzzy_match(char const * pattern, char const * str, int & outScore, uint8_t * matches, int maxMatches);
+}
+
+#endif // FTS_FUZZY_MATCH_H

src/listview_curses.cc

@@ -282,8 +282,14 @@ void listview_curses::do_update(void)
         }
 
         if (this->lv_show_bottom_border) {
-            mvwchgat(this->lv_window,
-                this->lv_y + height - 1, this->lv_x, width - 1, A_UNDERLINE, 0, NULL);
+            cchar_t row_ch[width];
+            int y = this->lv_y + height - 1;
+
+            mvwin_wchnstr(this->lv_window, y, this->lv_x, row_ch, width - 1);
+            for (int lpc = 0; lpc < width - 1; lpc++) {
+                row_ch[lpc].attr |= A_UNDERLINE;
+            }
+            mvwadd_wchnstr(this->lv_window, y, this->lv_x, row_ch, width - 1);
         }
 
         this->lv_needs_update = false;

src/readline_callbacks.cc

@@ -528,32 +528,43 @@ void rl_display_matches(void *dummy, readline_curses *rc)
     unsigned long width;
     __attribute((unused))
     unsigned long height;
-    int max_len, cols, rows;
+    int max_len, cols;
 
     getmaxyx(lnav_data.ld_window, height, width);
 
     max_len = rc->get_max_match_length() + 2;
     cols = max(1UL, width / max_len);
-    rows = (matches.size() + cols - 1) / cols;
 
     if (matches.empty()) {
         lnav_data.ld_match_source.clear();
     }
-    else if (cols == 1) {
-        lnav_data.ld_match_source.replace_with(rc->get_matches());
-    }
     else {
-        std::vector<std::string> horiz_matches;
+        string current_match = rc->get_match_string();
+        int curr_col = 0;
+        attr_line_t al;
+        bool add_nl = false;
 
-        horiz_matches.resize(rows);
-        for (size_t lpc = 0; lpc < matches.size(); lpc++) {
-            int curr_row = lpc % rows;
-
-            horiz_matches[curr_row].append(matches[lpc]);
-            horiz_matches[curr_row].append(
-                max_len - matches[lpc].length(), ' ');
+        for (auto match : matches) {
+            if (add_nl) {
+                al.append(1, '\n');
+                add_nl = false;
             }
-        lnav_data.ld_match_source.replace_with(horiz_matches);
+            if (match == current_match) {
+                al.append(match, &view_curses::VC_STYLE, A_REVERSE);
+            } else {
+                al.append(match);
+            }
+            curr_col += 1;
+            if (curr_col < cols) {
+                int padding = max_len - match.size();
+
+                al.append(padding, ' ');
+            } else {
+                curr_col = 0;
+                add_nl = true;
+            }
+        }
+        lnav_data.ld_match_source.replace_with(al);
     }
 
     tc.reload_data();

src/readline_curses.cc

@@ -62,6 +62,7 @@
 #include "ansi_scrubber.hh"
 #include "readline_curses.hh"
 #include "spookyhash/SpookyV2.h"
+#include "fts_fuzzy_match.hh"
 
 using namespace std;
 
@@ -71,6 +72,7 @@ static sig_atomic_t     got_winch   = 0;
 static readline_curses *child_this;
 static sig_atomic_t     looping      = 1;
 static const int        HISTORY_SIZE = 256;
+static int              completion_start;
 
 static const char *RL_INIT[] = {
     /*
@@ -79,6 +81,11 @@ static const char *RL_INIT[] = {
      */
     "set horizontal-scroll-mode on",
     "set bell-style none",
+    "set show-all-if-ambiguous on",
+    "set show-all-if-unmodified on",
+    "set menu-complete-display-prefix on",
+    "TAB: menu-complete",
+    "\"\\x0b\": menu-complete-backward",
 
     NULL
 };
@@ -237,6 +244,37 @@ char *readline_context::completion_generator(const char *text, int state)
                     matches.push_back(*iter);
                 }
             }
+
+            if (matches.empty()) {
+                vector<pair<int, string>> fuzzy_matches;
+
+                for (iter = arg_possibilities->begin();
+                     iter != arg_possibilities->end();
+                     ++iter) {
+                    const char *poss_str = iter->c_str();
+                    int score;
+
+                    if (fts::fuzzy_match(text, poss_str, score) && score > 0) {
+                        log_debug("match score %d %s %s", score, text, poss_str);
+                        fuzzy_matches.emplace_back(score, *iter);
+                    }
+                }
+
+                if (!fuzzy_matches.empty()) {
+                    stable_sort(begin(fuzzy_matches), end(fuzzy_matches),
+                        [](auto l, auto r) { return r.first < l.first; });
+
+                    int highest = fuzzy_matches[0].first;
+
+                    for (auto pair : fuzzy_matches) {
+                        if (highest - pair.first < 10) {
+                            matches.push_back(pair.second);
+                        } else {
+                            break;
+                        }
+                    }
+                }
+            }
         }
 
         if (matches.size() == 1) {
@@ -246,8 +284,8 @@ char *readline_context::completion_generator(const char *text, int state)
 
             last_match_str_valid = false;
             if (sendstring(child_this->rc_command_pipe[readline_curses::RCF_SLAVE],
-                           "m:0:0",
-                           5) == -1) {
+                           "m:0:0:0",
+                           7) == -1) {
                 _exit(1);
             }
         }
@@ -267,6 +305,7 @@ char **readline_context::attempted_completion(const char *text,
 {
     char **retval = NULL;
 
+    completion_start = start;
     if (start == 0 && loaded_context->rc_possibilities.find("__command") !=
             loaded_context->rc_possibilities.end()) {
         arg_possibilities = &loaded_context->rc_possibilities["__command"];
@@ -322,6 +361,19 @@ static int rubout_char_or_abort(int count, int key)
     }
 }
 
+int readline_context::command_complete(int count, int key)
+{
+    if (loaded_context->rc_possibilities.find("__command") !=
+        loaded_context->rc_possibilities.end()) {
+        char *space = strchr(rl_line_buffer, ' ');
+
+        if (space == nullptr) {
+            return rl_menu_complete(count, key);
+        }
+    }
+    return rl_insert(count, key);
+}
+
 readline_curses::readline_curses()
     : rc_active_context(-1),
       rc_child(-1),
@@ -377,6 +429,7 @@ readline_curses::readline_curses()
         stifle_history(HISTORY_SIZE);
 
         rl_add_defun("rubout-char-or-abort", rubout_char_or_abort, '\b');
+        // rl_add_defun("command-complete", readline_context::command_complete, ' ');
 
         for (int lpc = 0; RL_INIT[lpc]; lpc++) {
             snprintf(buffer, sizeof(buffer), "%s", RL_INIT[lpc]);
@@ -425,7 +478,9 @@ void readline_curses::store_matches(
         }
     }
     else {
-        rc = snprintf(msg, sizeof(msg), "m:%d:%d", num_matches, max_len);
+        rc = snprintf(msg, sizeof(msg),
+                      "m:%d:%d:%d",
+                      completion_start, num_matches, max_len);
         if (sendstring(child_this->rc_command_pipe[RCF_SLAVE], msg, rc) == -1) {
             _exit(1);
         }
@@ -503,10 +558,18 @@ void readline_curses::start(void)
 
                     SpookyHash::Hash128(rl_line_buffer, rl_end, &h1, &h2);
 
+                    if (rl_last_func == readline_context::command_complete) {
+                        rl_last_func = rl_menu_complete;
+                    }
+
+                    bool complete_done = (
+                        rl_last_func != rl_menu_complete &&
+                        rl_last_func != rl_backward_menu_complete);
+
                     if (h1 == last_h1 && h2 == last_h2) {
                         // do nothing
                     } else if (sendcmd(this->rc_command_pipe[RCF_SLAVE],
-                                       'l',
+                                       complete_done ? 'l': 'c',
                                        rl_line_buffer,
                                        rl_end) != 0) {
                         perror("line: write failed");
@@ -735,8 +798,10 @@ void readline_curses::check_poll_set(const vector<struct pollfd> &pollfds)
                 }
             }
             else if (msg[0] == 'm') {
-                if (sscanf(msg, "m:%d:%d", &this->rc_matches_remaining,
-                           &this->rc_max_match_length) != 2) {
+                if (sscanf(msg, "m:%d:%d:%d",
+                           &this->rc_match_start,
+                           &this->rc_matches_remaining,
+                           &this->rc_max_match_length) != 3) {
                     require(0);
                 }
                 this->rc_matches.clear();
@@ -783,6 +848,12 @@ void readline_curses::check_poll_set(const vector<struct pollfd> &pollfds)
                     this->rc_display_match.invoke(this);
                     break;
 
+                case 'c':
+                    this->rc_line_buffer = &msg[2];
+                    this->rc_change.invoke(this);
+                    this->rc_display_match.invoke(this);
+                    break;
+
                 case 'n':
                     this->rc_display_next.invoke(this);
                     break;

src/readline_curses.hh

@@ -203,6 +203,7 @@ public:
         return this->rc_highlighter;
     };
 
+    static int command_complete(int, int);
 private:
     static char **attempted_completion(const char *text, int start, int end);
     static char *completion_generator(const char *text, int state);
@@ -383,6 +384,19 @@ public:
         return this->rc_matches;
     };
 
+    int get_match_start() const {
+        return this->rc_match_start;
+    }
+
+    std::string get_match_string() const {
+        size_t space_index = this->rc_line_buffer.find(' ', this->rc_match_start);
+
+        if (space_index > 0) {
+            space_index = space_index - this->rc_match_start;
+        }
+        return this->rc_line_buffer.substr(this->rc_match_start, space_index);
+    }
+
     int get_max_match_length() const {
         return this->rc_max_match_length;
     };
@@ -408,6 +422,7 @@ private:
     std::string rc_line_buffer;
     time_t      rc_value_expiration;
     std::string rc_alt_value;
+    int rc_match_start{0};
     int rc_matches_remaining;
     int rc_max_match_length;
     std::vector<std::string> rc_matches;

src/view_curses.cc

@@ -516,7 +516,6 @@ void view_curses::mvwattrline(WINDOW *window,
         }
 
         if (attr_range.lr_end > attr_range.lr_start) {
-            string_attrs_t::const_iterator range_iter;
             int awidth = attr_range.length();
             int color_pair;
 

src/vt52_curses.cc

@@ -177,6 +177,9 @@ const char *vt52_curses::map_input(int ch, int &len_out)
             case 0x7f:
                 ch = BACKSPACE;
                 break;
+            case KEY_BTAB:
+                ch = BACKTAB;
+                break;
         }
         this->vc_map_buffer = (char)ch;
         retval  = &this->vc_map_buffer; /* XXX probably shouldn't do this. */

src/vt52_curses.hh

@@ -118,6 +118,7 @@ public:
     const static char BACKSPACE = 8;    /*< VT52 Backspace key value. */
     const static char BELL      = 7;    /*< VT52 Bell value. */
     const static char STX       = 2;    /*< VT52 Start-of-text value. */
+    const static char BACKTAB   = 11;
 
 protected: