Update performance section in alf man page.

3 files changed, 26 insertions, 18 deletions

diff --git a/TODO.txt b/TODO.txt
index e34e1d5..4da89d8 100644
--- a/TODO.txt
+++ b/TODO.txt
@@ -1,6 +1,6 @@
 - fuzz unalf and unalf with afl++ or similar.
 
-- clean up alf, try to make it faster.
+- clean up crunch.c.
 
 - figure out why unalf craps out when extracting files of 15MB
   but works fine for 14MB. when this happens, there's no error,
diff --git a/src/alf.1 b/src/alf.1
index 569e980..3733915 100644
--- a/src/alf.1
+++ b/src/alf.1
@@ -27,7 +27,7 @@ level margin: \\n[rst2man-indent\\n[rst2man-indent-level]]
 .\" new: \\n[rst2man-indent\\n[rst2man-indent-level]]
 .in \\n[rst2man-indent\\n[rst2man-indent-level]]u
 ..
-.TH "ALF" 1 "2025-11-29" "0.2.1" "Urchlay's Atari 8-bit Tools"
+.TH "ALF" 1 "2025-12-01" "0.2.1" "Urchlay's Atari 8-bit Tools"
 .SH NAME
 alf \- create Atari 8-bit ALF archives
 .\" RST source for alf(1) man page. Convert with:
@@ -195,14 +195,18 @@ compress/decompress such files, and you\(aqd have to have a hard disk and
 a DOS capable of handling multi\-megabyte files...
 .SS Performance
 .sp
-Performance is \fIhorrible\fP\&. This shouldn\(aqt be a real problem on
-modern multi\-GHz CPU, especially since most Atari 8\-bit files are
-small (usually under 64KB). Interestingly, it\(aqs not O(n^2), it scales
-linearly, O(1): Compressing a 1.3MB text file takes 0.7 seconds on the
-author\(aqs (rather modest) Intel i7 workstation, and a file 10x as large
-takes approximately 10x as long (7 seconds). A 50KB file is almost
-instantaneous, 0.05 seconds, which is more typical of the files you\(aqd
-actually use this with.
+Performance is pretty good, as of alftools\-0.3.0. For small files
+like you\(aqd use on an Atari (up to 50KB), it\(aqs basically instantaneous
+(under 0.02 seconds) on the author\(aqs modest i7 workstation. For a 1MB
+text file, it takes 0.05 sec; for 1MB of random garbage, it\(aqs 0.1
+sec (and the resulting ALF file is 36% larger than the garbage).
+.sp
+By comparison, \fBzip\fP takes 0.6 seconds to compress the 1MB text file,
+and 0.03 sec for the 1MB randomness (and the compressed file is still
+larger than the input, but only by 312 bytes). The speed demon is \fBarc\fP:
+it compresses the text file in 0.03s, and it\(aqs smart enough to \fInot\fP
+compress the random garbage (it uses the \(aqstore\(aq method, which \fBalf\fP
+doesn\(aqt have).
 .SS Timestamps
 .sp
 The date/time stamps stored in the archive are the \fBmtime\fPs of the
diff --git a/src/alf.rst b/src/alf.rst
index 4d48afc..0d2020e 100644
--- a/src/alf.rst
+++ b/src/alf.rst
@@ -170,14 +170,18 @@ a DOS capable of handling multi-megabyte files...
 Performance
 -----------
 
-Performance is *horrible*. This shouldn't be a real problem on
-modern multi-GHz CPU, especially since most Atari 8-bit files are
-small (usually under 64KB). Interestingly, it's not O(n^2), it scales
-linearly, O(1): Compressing a 1.3MB text file takes 0.7 seconds on the
-author's (rather modest) Intel i7 workstation, and a file 10x as large
-takes approximately 10x as long (7 seconds). A 50KB file is almost
-instantaneous, 0.05 seconds, which is more typical of the files you'd
-actually use this with.
+Performance is pretty good, as of alftools-0.3.0. For small files
+like you'd use on an Atari (up to 50KB), it's basically instantaneous
+(under 0.02 seconds) on the author's modest i7 workstation. For a 1MB
+text file, it takes 0.05 sec; for 1MB of random garbage, it's 0.1
+sec (and the resulting ALF file is 36% larger than the garbage).
+
+By comparison, **zip** takes 0.6 seconds to compress the 1MB text file,
+and 0.03 sec for the 1MB randomness (and the compressed file is still
+larger than the input, but only by 312 bytes). The speed demon is **arc**\:
+it compresses the text file in 0.03s, and it's smart enough to *not*
+compress the random garbage (it uses the 'store' method, which **alf**
+doesn't have).
 
 Timestamps
 ----------