Dialog Box

By Scott T Boyd, Editor

Challenging Challenge Challenged

One of the Huffman challenge entrants, Allen Stenger, found a bug in Bob Boonstra’s winning published Huffman code. It’s minor (ushort should be ulong) but it’s a real bug. Fixing it doesn’t change Bob’s speed. Here’s some background on the problem.

- Mike Scanlin
progchallenge@xplain.com

Subj: Huffman solution error?

I believe I’ve found an error in Bob Boonstra’s published solution to the Huffman challenge (MacTech, January 1995 issue). Look at p. 72, the definition of macro ProcessBitSlow. Observe that temp is declared as register ushort; contrast this with the immediately preceding macros ProcessBit and ProcessBitFast, where it is defined as register ulong. Actually all three should be ulong. The reason is that temp is the offset to the child node; since the node table could be as large as 256K bytes (64K nodes), temp can hold the offset in units of nodes but when ProcessBitSlow does the multiplication temp *= sizeof(DecodeNode), temp may overflow.

I was able to provoke this failure using the following example: The decoded numbers are either 3-bit numbers 0-7 or 16 bits numbers 8000-FFFF. The 3-bit numbers are encoded in 4 bits as 0-7 (i.e., a leading 0 bit followed by the number). The 16 bit numbers are encoded as themselves (they already have a leading 1 bit). Under this scheme A000 is decoded incorrectly.

- Allen Stenger.

Allen Stenger is correct in pointing out an error in the published solution for the HuffmanDecode challenge. As Allen points out, the declaration of the variable temp in macro ProcessBitSlow should be register ulong.

The bug shows up when both of the following are true: (1) the private storage available exceeds 64K, AND (2) numSymElems is large, greater than ~8K to ~32K, depending on maxMemoryUsage.

The bug does not occur if maxMemoryUsage is <=64K. Condition (2) for the bug is really when the size of the decode table exceeds the available storage. Since each node in the decode table is 4 bytes, and since there are (roughly) 2 nodes in the decode table for each element of the symbol table, the bug shows up under the following conditions:

numSymElems  maxMemoryUsage
> ~8K      > 64K
>~16K           >128K
>~32K           >256K (beyond Challenge limits)
max (2**16-1)   >512K (beyond Challenge limits)

(The relationship to numSymElems is approximate because it depends on the degree of balance in the tree representing the Huffman encoding.)

In these cases, some symbols will be decoded incorrectly. In the case cited by Allen, numSymElems is >32K, and the error shows up if maxMemoryUsage is >64K.

Correcting the bug changes a few xxx.W instructions to xxx.L, but shortens the code generated by the macro in question by one instruction, and actually improves the run time by an almost imperceptible amount.

It just goes to show you can’t do enough testing. My testing used Huffman encodings that were generated from large text files, so I got encodings with a few thousand elements. Still, I should have caught this one by inspection. Congratulations to Allen for detecting the error, and good luck to him in future Challenges.

- Bob Boonstra

Anonymously Yours

I’ve heard that it’s possible to anonymously send e-mail to someone. I’ve also heard that in order for this to be really anonymous you need to send it through three or four anonymous re-mailers. Could you publish the addresses of these remailers and give an example of how to send e-mail that has a very high chance of remaining anonymous?

[name and address withheld]

Great Prograph Coverage

Your January ’95 issue looks like one of the best ever. Two CD-ROMs, who could ask for more? I was also very excited to see Kurt Schmucker’s article on Prograph Commands. I hope that these articles on Prograph become a regular feature. As a former member of the Software Frameworks Association, I’m glad to see MacTech really taking up the slack where Frameworks left off. I have been working with Prograph CPX for several months now and find it to be a great tool.

Please keep up your coverage of it, and other new OOP technologies.

- Steve Wilson , Emergent Behavior
emergent@aol.com

Do the Cobbler’s Children Always Go Barefoot?

I am inspired to write after reading the letter by Steve Weller, "A Trail of Good Intentions", printed in the December Dialog Box. Like Steve, I always thought that "C" stood for cryptic but C++ is much worse.

I can best describe myself as a classic Mac hacker, that is one who programs the Mac for the sheer joy of it and not

to make a living. As a programmer, my user interface is the editor I use. I have made a few dollars on shareware but mostly I program the Mac just because I love the Mac’s user interface.

You can judge the low quality of the editors that come with most C++ compilers by the number of third party editors being sold. I personally can’t stand to use any of the popular C++ environments (with or without third party editors) because they don’t just compare to the editor in THINK Pascal.

THINK Pascal’s editor automatically formats my source code. It finds syntax errors before I waste time trying to compile. Most of the Mac user interface is in the Runtime and Interface libraries so I seldom have to include other libraries or interfaces. The stops are easier and quicker to use than the ones in C++. Best of all, the compiler and linker are very fast for reasonable sized programs.

The only thing I have found that really slows THINK Pascal is when I play with MacApp, TCL or one of the other bloated object libraries. For my serious programs I use THINK Pascal with a non-object shell generated by Marksman. In spite of the fact that THINK Pascal hasn’t been updated for several years, I am still more productive using it than C++.

I wish the technical support engineers who write the Symantec Top Ten would answer these two questions.

#1 When is Symantec going to produce a new version of THINK Pascal that will generate both 680x0 and PowerPC code?

#2 Why don’t they transplant the advanced editor, compiler and linker technologies found in THINK Pascal to their slow user hostile C++?

- Fred Johnson
70651.3171@compuserve.com