From: Steinar H. Gunderson <steinar+git@gunderson.no>
Date: Sat, 3 Oct 2020 08:49:10 +0000 (+0200)
Subject: Fix searching for very short (1 or 2 bytes) queries.
X-Git-Tag: 1.0.0~63
X-Git-Url: https://git.sesse.net/?a=commitdiff_plain;h=d5f6c3c0a4419f48ac62beca86e28c9e6aea2412;p=plocate

Fix searching for very short (1 or 2 bytes) queries.

plocate had assumptions about the layout of the file, that no longer
held. Use the pad field to simplify things.

This requires a database rebuild, but only for short queries.
Normal queries will continue to work, so there's no version bump.
---

diff --git a/db.h b/db.h
index cb0a7a8..ca8b3ea 100644
--- a/db.h
+++ b/db.h
@@ -8,7 +8,7 @@ struct Header {
 	uint32_t version;  // 0.
 	uint32_t hashtable_size;
 	uint32_t extra_ht_slots;
-	uint32_t pad;   // Unused.
+	uint32_t num_docids;
 	uint64_t hash_table_offset_bytes;
 	uint64_t filename_index_offset_bytes;
 };
diff --git a/plocate-build.cpp b/plocate-build.cpp
index 658aea2..c680449 100644
--- a/plocate-build.cpp
+++ b/plocate-build.cpp
@@ -377,6 +377,7 @@ void do_build(const char *infile, const char *outfile, int block_size)
 	}
 	corpus.flush_block();
 	dprintf("Read %zu files from %s\n", corpus.num_files, infile);
+	hdr.num_docids = corpus.filename_blocks.size();
 
 	// Stick an empty block at the end as sentinel.
 	corpus.filename_blocks.push_back(ftell(outfp));
diff --git a/plocate.cpp b/plocate.cpp
index fa70f76..fc8c4be 100644
--- a/plocate.cpp
+++ b/plocate.cpp
@@ -188,12 +188,7 @@ void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string)
 
 size_t Corpus::get_num_filename_blocks() const
 {
-	// The beginning of the filename blocks is the end of the filename index blocks.
-	uint64_t end;
-	complete_pread(fd, &end, sizeof(end), hdr.filename_index_offset_bytes);
-
-	// Subtract the sentinel block.
-	return (end - hdr.filename_index_offset_bytes) / sizeof(uint64_t) - 1;
+	return hdr.num_docids;
 }
 
 size_t scan_file_block(const vector<string> &needles, string_view compressed,