From 7ea04be9ee25d1a768a67737a1c49a4678bd61f5 Mon Sep 17 00:00:00 2001
From: Anders Holck <anders@holck.se>
Date: Thu, 30 Apr 2026 21:20:05 +0200
Subject: [PATCH] Add client-side language pre-filter to reduce false
 translations

- Common Swedish/English words checked before API call
- Word boundary matching (non-alpha boundaries)
- Lower threshold for short messages (1 hit vs 2)
- Only filters languages listed in ai_skip_langs
---
 main.c | 40 +++++++++++++++++++++++++++++++++++++++-
 1 file changed, 39 insertions(+), 1 deletion(-)

diff --git a/main.c b/main.c
index c4ac29f..daa42bb 100644
--- a/main.c
+++ b/main.c
@@ -188,11 +188,49 @@ static void ai_config_load(void)
 static int needs_translation(const char *text)
 {
 	if (!ai_cfg.enabled || !translate_enabled) return 0;
-	/* Skip very short messages (nicks, URLs, single words like "ok") */
+	/* Skip very short messages */
 	int words = 0;
 	for (const char *p = text; *p; p++)
 		if (*p == ' ') words++;
 	if (words < 1 && strlen(text) < 6) return 0;
+
+	/* Pre-filter: if text contains common words from skip languages, skip */
+	static const struct { const char *lang; const char *words[23]; } lang_words[] = {
+		{"swedish", {"jag", "och", "att", "det", "inte", "var",
+		             "som", "för", "med", "har", "den", "kan",
+		             "ska", "till", "eller", "men", "där",
+		             "när", "från", "ett", "en", "ta",
+		             NULL}},
+		{"english", {"the", "and", "that", "this", "with",
+		             "have", "was", "are", "you", "not",
+		             "from", "but", "for", "can", NULL}},
+		{NULL, {NULL}}
+	};
+	char lower[1024];
+	snprintf(lower, sizeof(lower), "%s", text);
+	for (char *p = lower; *p; p++)
+		if (*p >= 'A' && *p <= 'Z') *p += 32;
+	int hits = 0;
+	for (int i = 0; lang_words[i].lang; i++) {
+		if (!strstr(ai_cfg.skip_langs, lang_words[i].lang)) continue;
+		for (int j = 0; lang_words[i].words[j]; j++) {
+			const char *w = lang_words[i].words[j];
+			size_t wlen = strlen(w);
+			char *p = lower;
+			while ((p = strstr(p, w)) != NULL) {
+				/* Check word boundaries (non-alpha on both sides) */
+				int before_ok = (p == lower) ||
+					!(*(p-1) >= 'a' && *(p-1) <= 'z');
+				int after_ok = !p[wlen] ||
+					!(p[wlen] >= 'a' && p[wlen] <= 'z');
+				if (before_ok && after_ok) { hits++; break; }
+				p++;
+			}
+		}
+	}
+	int threshold = (words < 4) ? 1 : 2;
+	if (hits >= threshold) return 0;
+
 	return 1;
 }