Developing for fun with OS/2 and OpenWatcom

By Axel Meiss

For more than a decade I have been using OS/2 now. I got used to it when I had to use OS/2 2.1 when I got employed in my first IT-company who chose OS/2 as their platform to develop software for warehouse systems. Back then, we used the 16-bit compilers MS C 6.0 and MS Cobol 4.5. (yes, I really did Cobol, and it's not as bad as its reputation). Having used only the Atari ST and MS-DOS 5, I was quite impressed about a real multitasking system with a nice GUI. But, I was also impressed by the possibility of having multiple character mode applications in multiple windows. I wasn't used to character mode applications because the Atari ST didn't provide that, and DOS only allowed me to run it all full-screen. Though I had an Unix machine available, I didn't like the terminal access with that machine. OS/2 was much more convenient. I upgraded my machine at home with additional 4 MB of RAM, so I could run OS/2 2.0 (a copy which I had bought inexpensively before) at home with 5 MB.

Commercial compilers were expensive then ($400+), so I was glad that there was a free compiler available, the EMX-port (by Eberhard Mattes) of the GNU compiler. I didn't develop much at home because I was already developing in the company. But, it helped me a lot to explore OS/2. Later, I purchased a copy of Watcom C/C++ 10.0, my first commercial compiler because they gave it away for only $200 or so. I liked the idea of a multi-platform compiler which supported OS/2, DOS, Windows and even Novell. In the company, we were evaluating VisualAge C++ 3.0 for OS/2 which was quite clumsy, compared to the Watcom compiler. I was quite annoyed by the fact that VisualAge C++ didn't use the standard LINK386 because it took me some time to understand it and the new linker also created Extended Attributes for my executables which I didn't like. But, I got used to it...

Dynamic Linking
One of the interesting new things then was the concept of Dynamic-Link-Libraries(DLL). First regarded as a totally cool way of creating applications it has become a source of potential trouble, especially in the way of different versions of a DLL. This problem is known under Windows as the "DLL-Chaos". Some programs require a specific version of a DLL and will not run with an earlier version. If now a program ships with the older version the installation may replace the newer DLL with the older one. It can be very hard to keep track which is the most recent version of the DLL. OS/2 is by no means immune to the DLL-Chaos, so a lot of care must be taken when DLLs should be shipped with an application. When creating DLLs one also should watch the naming of a DLL. Actually, under OS/2 a DLL name it is advised to be unique, i.e. you should not have two different DLLs which are named both "SOMEDLL.DLL" which are located in different directories. Whenever one of those DLLs is loaded by an application, an application that requires the other DLL may not run properly because it refers to the DLL loaded earlier. You can avoid this problem by explicitly loading a DLL with fully qualified file name (like C:\SPECIAL\SOMEDLL.DLL). But this requires that the calling application "knows" where the DLL is located. Reportedly, you can also force an explicit load of some version of a DLL in command line sessions with SET LIBPATHSTRICT=T even if some other version of the same DLL is already loaded (which was introduced with some later kernel of Convenience Package 1 or so). It seems to me though that this was included for testing DLLs purpose only. Usually, a DLL is placed in some directory listed in the LIBPATH statement of CONFIG.SYS. The naming problem was even a little aggravated with OS/2 Warp 3 when DLL file names were restricted to follow the 8.3-file-naming convention to allow the DLL to reside on a FAT-formatted volume.

Though the namespace for a DLL is quite large (> 36^8) it's very easy to create conflicting DLL names when there should be "speaking" names like "FILEIO.DLL" or "HELPER.DLL" instead of "CY100A9L.DLL". Of course, one of the solutions is to avoid DLLs at all. Very often, this is the best solution. Sometimes though, it is sensible to create and use DLLs, especially if those DLLs provide additional functionality. Take Impos/2 e.g. which uses a separate DLL for each graphics format. In case there are new file formats, only a new DLL is needed while the main application doesn't have to change. As for me, I have come to the opinion that not putting much functionality into a single DLL is the best way when creating DLLs because the more code a DLL contains the more errors are likely to exist and the more often a new version of a DLL may be needed.

Just giving a short summary what has been explained in many articles about OS/2 before: OS/2 has two ways of handling DLLs: Load-Time Linking and and Run-Time Linking Load-Time Linking means the DLL is needed when a program starts, otherwise the program doesn't start at all and the error SYS1804 is produced. This happens when you have installed Mozilla 1.7 without having the LIBC0x.DLL somewhere in a directory which is listed in your LIBPATH. Run-Time Linking means the program starts, then the DLL is loaded by the program "manually" requiring handling by the program. Obviously, the latter method is more flexible and is widely used by key applications like web browser Mozilla. Even the Cobol environment I used at work took advantage of Run-Time Linking.

Why OS/2 ?
When telling people that I use eComStation I usually get asked "What is that ?". When I tell them it's a system based on OS/2 people start staring at me. I frankly mention that I was tired of all the security updates to Windows NT 4.0 which I used for internet access until three years ago. Of course, I had still a box with OS/2 Warp 3 Connect installed at home. I decided to purchase a copy of eComStation 1.0, mainly to run Mozilla. Though I had a package of Warp 4 at home I never installed it for production because I was still satisfied with Warp 3. Getting eComStation was a smart decision and I later ordered version 1.1 as an upgrade to my Warp 4 package. Other people who are known to OS/2 asked me why I chose that platform for development. They remark that Linux offers much more software. In fact, that is correct. Yet, there is much more to do for OS/2.

However, you are still at risk to develop and publish something that's not needed anymore because it's already there. This goes especially for easy tasks and small programs like utilities. There are only few reasons to publish something which is already covered by some other program, e.g. Actually, with OS/2 there are too many tasks that a developer can choose from. But most of those tasks are very complex and require a lot of effort to implement them. The program I explain here is very small and hardly worth mentioning unless for demonstrating some programming technique. As far as larger projects are concerned, I strongly support the idea of the bounties at OS/2 World.com where users are encouraged to donate for potential projects. In the past, I have purchased a sponsor unit for Netlabs to support the great work that is done there. I plan to purchase some more.
 * the new program overcomes a severe limitation of the existing solution
 * the new program includes the source while the old doesn't
 * the new program is a ported solution from other systems

Why OpenWatcom ?
Back in 2003, OpenWatcom 1.0 became available and being a long-term user of Watcom C/C++, I ordered the media quite soon to support the project. It didn't really change in handling from previous versions, so I knew instantly how to operate the IDE. I was quite pleased by that. That is the main reason why I use it. Being the only Open Source compiler for OS/2 besides GCC, it deserves continuous usage by developers. Though OpenWatcom is available for free download, I encourage all users interested in development to order the OpenWatcom CD. As far as I know, OpenWatcom is the only supported C++ compiler left that creates 16-bit executables both for DOS and OS/2. VisualAge C++ 3.0 for OS/2 is not supported any more yet it is still very common though. That's one of the reasons why I used it for compiling MD5SUML though subsequent projects were created with OpenWatcom. We mustn't forget though that many larger projects have become possible only be the availability of the GNU C++ Compiler.

MD5SUML
MD5SUML was a test project to see how those file APIs introduced with Warp Server for e-Business worked which made it possible to create and access files with a size over 2 Gigabytes. It was also tested how to compile a program without 64-bit integer support of the compiler (VisualAge C++ 3.0) and without the Warp 4.5 toolkit available. Interestingly, by then (February 2004) at Hobbes there was no message digest program available for handling large files. I was quite puzzled by that. Following the rule of overcoming a severe limitation of the existing solution, I uploaded the whole stuff to Hobbes as it would be valuable to someone though MD5SUML has only a very limited functionality, i.e., calculating a large file's MD5 sum.

The program itself was yet another implementation of Ron Rivest's reference code of MD5 in RFC1321. The Message Digest 5 is a cryptographic hash algorithm which allows files to be verified against tampering. Usually, it is used as a one-way hash to encrypt passwords. It is encryption because the password cannot be reconstructed from its MD5 sum. This feature of cryptographic hashes is the "one-way function". To verify a password, the MD5 sum of the password is calculated and is then compared to the MD5 sum of that user's password. If they compare equal the password is accepted. So, on the authorization side, no user passwords are stored but only their MD5 sums. However, it may be that different passwords produce the same MD5. That is called a "collision". A strong message digest requires that collisions cannot be produced except by brute force. To verify files against accidental change the good old Cyclic-Redundancy-Check is usually preferred.

Open Source!
Of course, I uploaded the source, too so anybody with experience could modify the program to fit his own needs.

But there is a more serious reason for uploading the source, too: By providing the source, you enable others to validate that the code is free from "hidden" features. The user doesn't have to trust you any more, he can see for himself if the code only does what it is supposed to do. Obviously, it can be embarrassing to publish poor code. But it's better to include the source even if you are not sure about your code.

The successor to MD5SUML
In 2004, there were reports that collisions could be found with MD5 sums more easily than probability would expect. A collision means finding two files which produce the same MD5 sum. Though it doesn't affect file integrity I began to check for other message digests. I encountered the Java program JACKSUM (www.jonelo.de). I was quite impressed by the number of different message digests it can calculate. To my surprise, I didn't find any stand-alone program for OS/2 that calculates an SHA-1 checksum. I realized that a native implementation of SHA-1 would be a significant enhancement to MD5SUML. Of course, extending MD5SUML by SHA-1 leads to the question why only SHA-1? What about other message digests and how should it be handled then?

A new executable like CHCKSUML.EXE each time a new message digest is implemented thus creating versions from 1.0 to 19.0?

Creating different programs like MD5SUML.EXE for MD5, SHA1SUML.EXE for SHA-1 etc.?

It was clear that OS/2 offered something much better. As I mentioned earlier, OS/2 offers the concept of Dynamic Linking. So, the better solution was to create a different DLL for each of the different message digests being processed by a frame program. It became quite obvious that Load-Time Linking was not appropriate. Load-Time Linking requires that all needed DLLs are present at program start-up. So, each time a new message digest is implemented it requires the original application to be linked again. That is what you need an Import Library or a DEF-file for. It's actually as good as creating a new executable program when a new message digest should be implemented. But with Run-Time Linking you don't have to run the linker again. The main application stays the same while new DLLs can be added. Of course, you have another problem then: can all the message digest functions be called the same way ? After investigating, I came to the conclusion that that should be possible. Each message digest DLL has then the following functions: Though they all digests perform different operations, the message digest algorithm must be transformed to fit that interface. Later, I will give an example here. Those functions above will be exported functions of the DLL. You can either create a DEF-file or you tell the linker to export those functions. With MD5SUML I have created a Module Definition file (MD5SUML.DEF) which tells the linker where to resolve the API-function DosOpenL. With OpenWatcom I chose the latter option.
 * DIGESTINIT performs necessary initialization
 * DIGESTUPDATE performs the calculation of the input stream
 * DIGESTRESULT performs final steps and creates the output

The new program
MD5SUML could only be used by Warp 4 users who had at least Fixpak 13 installed. That was ok because the others could still use MD5.EXE (md5_os2.zip). As I didn't find a program for calculating SHA-1 sums I considered it a must not to exclude Warp 3 users from usage. I later found that there was a REXX script which calculates SHA-1 (Daniel Hellerstein's rexx_md5.zip) but the information about SHA-1 was really hidden there. Though Warp 3 users are not able to check large files they can, with the new program, calculate different checksums, too. It would have been a pity if the message digest program was only available to those who used the Warp 4.5 kernel. I apologize for not creating 16-bit executables which are needed for OS/2 1.3 users. The easiest way to support both Warp 3 and Warp 4.5 was to create two different programs which only have different file access routines. Of course, they use the same message digest DLLs. It gave me a minute of trouble finding the name for the frame program but DIGEST was a proper name. So, as there should be two different programs, I chose
 * DIGEST.EXE - for Warp 3 and Warp4 before Fixpak 13
 * DIGESTL.EXE - for Warp 4 with Fixpak 13, Warp 4 Convenience Package and eComStation

As a matter of course, DIGEST.EXE can be run by eComStation and Warp CP, too.

DIGESTL.EXE is mainly an adoption of what was developed in MD5SUML with the message digest calculation being done outside the EXE in the DLL part.

Usage of DIGEST
The program is run on command line like [c:\]digestl case is a normal file or '-' when input is taken from STDIN is the name of the message digest algorithm [case] is optional and will produce the message digest in lower case letters The key enhancement is supplied by mandatory parameter which says which DLL is used. DIGESTL will prefix that parameter with a 'D' and will then try to load the corresponding DLL according to the LIBPATH statement in CONFIG.SYS, e.g. "sum64" as becomes DSUM64.DLL.

I chose composing the DLL name that way because it might be that someone else ships some DLL some time whose name is MD5.DLL. Later I found that there is already an MD5.DLL as part of the MIT/GNU Scheme package. Nevertheless it is only a weak protection against duplicate DLL names but for easy handling of the program I didn't dare to use cryptic names like CY100A9L.DLL. On my system I have put the method-DLLs into C:\OS2\APPS\DLL which is included in the LIBPATH statement of CONFIG.SYS by default.

Last year, I was written to by a user who needed to check his DVD-images. The tool he used could write the content of the image to STDOUT so he asked me if MD5SUML could take input from STDIN. Though MD5SUML can do so it produces wrong results. I created the custom program MD5STDIN for him which allowed him to check the stream from STDIN. When he asked me if I could publish MD5STDIN I told him that I would include that feature with DIGEST because DIGEST is the more flexible program. So, if you need to check STDIN stream you simply use the dash (-) as a file name. It may be of interest that DIGEST.EXE which doesn't support large files can check large files though when redirecting a large file to STDIN.

I was also asked why both MD5SUML and DIGEST produce output in upper case. It has been a tradition that hexadecimal numbers are printed in capitals to stress that those are numbers and not regular texts. babe2feed may more easily be taken for a phrase than BABE2FEED However, I have added the option to print the message digest in lower case by the optional first parameter "case" because it was requested. I dropped the optional formatting of the output as it is in MD5SUML (1234-5678-...) because it didn't seem to be used at all.

Now, it may be remarked that this program should have been one that makes use of the GUI. This is seriously considerable. As I wanted a command line program it would have been needed to create an extra project, let's call it PMDIGEST. Displaying the result while not extending the functionality would look like this:

This is not what I consider well done. So, it's better to stick with the command line program only. Of course, there is a PM program available for calculating MD5 sums: SigmaMD5 by Keith Merrington.

Back in July last year, when SigmaMD5 was first uploaded to Hobbes I contacted the author saying it may be sensible that he did the PM version while taking advantage of the sample source concerning large file support provided with DIGEST. Unfortunately, the program doesn't come with source, so I don't know if he did. The nice thing about SigmaMD5 is that it loads the DLL for handling large files according to the version of OS/2. Nevertheless, SigmaMD5 has a different objective than DIGEST. Its main purpose is comparing two directories or drives like Roman Stangl's XCOMP2 while DIGEST is more or less sample code doing something useful.

Let's make it
Now, we have come to the section where it is shown how to create a simple message digest DLL. I have chosen an algorithm that simply adds all the bytes of a file and produces the sum of it as a 64-bit value. OpenWatcom supports the use of 64-bit integers unlike VisualAge C++ 3.0. It is required that OpenWatcom has been installed on the system. For good reason, I have not included the source in the DIGEST package because I want the user to really compile the source himself. The following C source code must be put into a file named SUM64.C by copy & paste. It is preferred to create a directory E:\SUM64 where SUM64.C can be saved and which will serve as project directory. /* SUM64.C -- snip start                                             */ /* SUM64.C Copyright (c) 2006 Axel Meiss                             */ /* may be distributed freely as long as copyright notice is preserved */


 * 1) include 
 * 2) include 
 * 3) include 

unsigned long long sum64; /* OpenWatcom has 64-bit integers */

int DIGESTINIT(void) { sum64 = 0; /* the sum is initialized to zero */ return 0; }

int DIGESTUPDATE(unsigned char *buffer, unsigned long ulSize) { unsigned long ulIndex; for(ulIndex = 0;ulIndex < ulSize;++ulIndex) sum64 += buffer[ulIndex];

return 0; }

int DIGESTRESULT(unsigned char *output) { int len; int longlen; int shiftsize; unsigned char *buffer = output; unsigned int iValue; char digits[16] = {'0','1','2','3','4','5','6','7','8','9','A','B','C','D','E','F' };

longlen = sizeof(sum64) - 1; for(len = 0;len <= longlen;len++) {   shiftsize = (longlen - len) * 8; iValue =  (sum64 >> shiftsize) & 0xff; *buffer++ = digits[iValue >> 4]; *buffer++ = digits[iValue & 0xf]; } *buffer = '\0';

return 0; } /* SUM64.C -- snip end -- */ Now, that the source has been saved to SUM64.C we can run the OpenWatcom IDE by opening the OpenWatcom folder and run OpenWatcom prompt and type "start ide". We create a "New project". The following window will appear. The user may excuse that the German version of the File Selector box is shown

After we have done that all we press function key F5 and let OpenWatcom do all the processing. Now, if everything went fine, we have DSUM64.DLL which we test by issuing the command [e:\sum64]digest sum64 sum64.c I may be asked why I chose a hexadecimal string as output instead of binary data. The reason is that different message digests have different sizes. SUM64 has 64-bit size, MD5 has 128-bit size and SHA-1 has 160-bit size. Strings in C are limited by the NUL character so its length is calculated by regular string functions. Otherwise, I would have to provide some length information somewhere. This way, the calling application doesn't have to know about the size of the message digest as long as the output buffer is large enough to hold the string with terminating NUL character.

Using DIGEST in REXX scripts
Sometimes, it can be useful to create the checksums of a whole list of files. When I created a REXX script to produce those checksums I found both in MD5SUML and DIGEST the serious flaw that a non-zero return code is not supplied when DIGEST tries to open an inaccessible file. This error has been corrected for DIGEST with the actual version. The following REXX script lists all files of a directory and its subdirectories and their respective 64-bit sums: /* files digest */ call RxFuncAdd 'SysLoadFuncs', 'RexxUtil', 'SysLoadFuncs' call SysLoadFuncs parse arg filenames say filenames if filenames == "" then filenames = "*"

call SysFileTree filenames, 'files', 'FOS' do i = 1 to files.0 fname = '"'files.i'"' digestl sum64 fname ">> list.s64" if rc <> 0 then do        zeile = 'Error' rc  fname commando = "@echo " zeile " >> list.s64" commando end end Using such REXX script can be handy when you have only a command line available and want to create a list of files with their checksums.

Things to do
When you check the JACKSUM homepage you find that there are a lot of different message digests that are included with JACKSUM. DIGEST only ships with two (MD5 and SHA-1). It is a task then to implement some of the remaining message digests like RIPEMD-160 which I plan to do soon. Though there are enough programs for calculating CRCs at Hobbes (Radim Kolar's crc.zip e.g.) CRC should be implemented, too. Another task may be the creation of a REXX callable DLL like RXDIGEST.DLL.

Acknowledgements

 * Johann Nepomuk
 * Ingo Steiner
 * Jeramie Samphere

(c) 2006 by Axel Meiss

''Axel Meiss is a long time user of OS/2 since OS/2 v2.1. He serves as Vice-President and Director at the OS2eCS Organization.''