Using The GNU Configure Scripts In OS/2

From EDM2
Jump to: navigation, search

Written by Dale DePriest


This article describes the ability to use GNU configure scripts within the OS/2 environment. These configure scripts are a key tool in permitting the automatic configuration of source code among a wide divergence of Unix systems. Extending their use to OS/2 opens the possibility of much more high quality and easily available software than we might otherwise enjoy. People have been porting Unix applications to OS/2 for years and this article hopes to help continue this practice by making the task a bit easier.

Of course if you are using a Unix script to configure and port Unix applications and utilities you might expect that the tools would depend upon a Unix environment. And this is indeed the case. Luckily pioneers in this effort have already developed most of the tools needed to establish a Unix environment for you. Names like Eberhard Mattes (The E and M of EMX) and Kai Rommel have already built many of the tools that you may need. Eberhard has supplied and continues to enhance a Unix-like linkable shared library layer plus a modified version of the GNU C compiler that supports most of the syntax and semantics of the Unix C language. While EMX supplies GCC as part of the installation there is no reason to believe that it requires GCC in order to work. EMX itself is a set of libraries that could be linked to any C or C++ compiler. Kai Rommel has used EMX to port many of the Unix utility commands into the OS/2 environment as well. These commands are needed both by the configure scripts and by many of the Unix makefiles. Many other people have contributed to this effort as well and many Unix-like commands are available from other sources.

Using The KSH Shell

In order to run the configure script you will need a Unix shell. This article describes the KSH shell currently released on hobbes and many other FTP sites. It is freely available and you can also obtain the source if you wish to build a modified copy for you own use. Unlike many of the earlier porting efforts of Unix code the OS/2 version is fully integrated into the source thus providing you the assurance that when the source changes, the OS/2 variants will be carried forward. The sources themselves are called pdksh, the public domain korn shell, and implement a very good emulation of the original AT&T Korn shell plus many POSIX enhancements. The configure script itself, however, does not require any Korn shell features but rather depends on the older Bourne shell syntax and semantics, of which a Korn shell is a superset. You should be able to use any good implementation of a Bourne shell to run this script.

The shell itself can be called any name you wish. Its behavior does not vary based on the name unlike some other Unix shell programs. To run the configure script it should be named sh.exe. Using the sources you can build many different variations of the shell using options in the config.h file. A full-blown Korn shell can be built or a complete subset that behaves very much like a Bourne shell. The smaller version is excellent for script support. In the run time release I have provided the full shell under the name ksh.exe and the smaller version under the name sh.exe.

Configure scripts assume the presence of Unix commands so you will need to insure that you have the correct selection available for the shell and that they are in your current search path. If you get command not found errors when running the script you will need to find (and download if necessary) a copy of the appropriate command. There are several Unix commands that have exactly the same names as OS/2 commands so be sure the Unix versions are earlier in the search path.

In Unix an executable is tagged with an attribute to make it known to the system as an executable. In OS/2 this is done with an extension. This can lead to unique problems when running the configure scripts. KSH knows all of the standard OS/2 extensions plus .ksh and .sh; and, as a special feature, it can even run commands that do not have any extensions. Most often commands in Unix do not have any extensions and scripts either have no extension or use the name of the shell as an extension. This is only a Unix convention, not a requirement, and to execute a Unix command you must type the exact filename. In OS/2 the .ksh or .sh extension tells the KSH that this is an executable shell script. A file without an extension may be either a script or a binary executable. Scripts need not necessarily be a KSH shell scripts however. The standard Unix #! line at the top of the file determines the actual command that will be used to run the script. Should the same filename with a different extension exist in the same directory KSH will use the extension specified in the script or if no extension is specified then the search order is, no extension, .ksh, .sh, .cmd, .exe, .com, .bat. Note that if the script specifies a different extension than one of the ones listed it will not be recognized as an executable by KSH.

Running Unix GNU Configure Scripts

KSH can be used to run GNU configure scripts, however, I can't guarantee that the results will match the true configuration for OS/2 since this is dependent on the way the individual configure scripts are written. You should expect that you may have to hand modify the configure script, the results of the configure, or at the very least check the results to insure that they look reasonable. Even in Unix itself the scripts may not always correctly identify a particular idiosyncrasy of a particular hardware platform configuration.

To set up to run configure scripts you will need to have a copy of KSH called sh.exe somewhere in your path and, of course, you should be running within a KSH shell. One of the very early things done by most configure scripts is to try and figure out the compiler. This will usually fail since it generally depends on setting the IFS (To understand why, see discussion in the KSH readme.os2 file on colon versus semicolon as a path separator). To work around this problem simply set CC to the compiler you want to use prior to running the script. For example to use gcc use the command: 'export CC=gcc'. Most configure scripts couldn't figure out the names of OS/2 compilers anyway. Now you should be able to simply type 'configure' and watch it work. If it seems to fail right off the bat and can*t seem to find any commands look in the script and see if PATH is being set by the script. If it is, please just delete that line and try again.

At some point the configure script will build a config.status script to perform the final step. If this is not run using 'sh -c config.status' or some such command in configure, then it will fail since .status is not an OS/2 legal executable suffix. You can run it manually however by typing 'ksh config.status'. This is the command you should use if you need to run config.status later. If you have problems you should inspect the config.status script and fix it as required. Using the command: ksh -x config.status will echo the lines as they are executed which should aid in debugging. Running config.status is a good way to reset any of the changes you may have made to config.h or the Makefile.

Many configure scripts actually build and execute small programs to test for certain features of the operating system or compiler. In OS/2 using EMX, the final link step for a program must have the .exe extension tied to the executable name or the linker will not link correctly. A few configure scripts already know about this requirement and will work OK, but, unfortunately most do not. You may need to modify the configure script to insure the correct executable name. Once the program is built it can be renamed to a name without an extension if KSH is being used to run it since often the script also checks to see if a file with this name was created by the compiler. A line similar to the following may work for you if you are using a version 2.0 configure:

ac_link='${CC-cc} -o conftest.exe $CFLAGS $CPPFLAGS $LDFLAGS \
conftest.$ac_ext $LIBS && mv conftest.exe conftest'

Figure 1) Modification for using Configure 2.0

This trick takes advantage of KSH's ability to run a command without an extension. Even with these changes you may still find that configure does not correctly identify some OS/2 features. Many configure scripts reuse a line like ac_lnk many times over the course of running the program, however some scripts use unique names each time they build a test program. For these configure programs it may be impractical to modify the script. EMX has a feature that permits it to attempt to fake this requirement for makefiles and configure scripts. By setting a compiler variable (usually LD, see EMX documentation for details) to include -Zexe you can force the link step to build an executable with the .exe extension on it and to create an empty file with just the name. This empty file is meant to satisfy Unix makefiles and may fool configure scripts as well. I say *may* because some configure scripts not only check for the existence of the executable name but also expect it to have size! These configure scripts wont be fooled.

Unfortunately the KSH ability to run a file without an extension can cause a conflict with the EMX feature to automatically build two files with the same name but a different extension. KSH will try and execute the dummy file and will fail. There are several ways correct this problem. If you know the name of the file then an easy solution is to define an alias for KSH use. For example the command *alias prog=prog.exe* would always execute prog.exe even if a file named prog were present. For permanent use this alias could be defined in your KSH startup file. Use of an alias is supported by both KSH and SH but is not exported to a script. Aliasing must be reestablished in each child. Should a more permanent or automatic solution be desired, I would recommend getting a copy of the KSH source file and customizing the exec.c file to alter the search order. A working example of a search order for configure scripts using the -Zexe option to EMX might be (where . represents a file name without an extension): .ksh, .exe, ., .sh, .com, .cmd, .bat. You may prefer .com ahead of .exe. While this search order wouldn't be intuitive to interactive use it does have merits for scripts. It is also possible to modify KSH so that zero length files without an extension would be ignored, and, if there is sufficient interest, I could modify the next version.

Building The Executable

Once you have gotten configure to successfully run, you should find a config.h file containing the platform customization flags and a makefile in the directory containing the sources. You may need to edit the config.h file either now or as a result of errors that you define during your compilation attempt. In addition you will probably need to edit the Makefile. Makefile edits depend somewhat on which make command you use. I usually use either make from GNU or nmake from IBM. Both are available from hobbes. At the very least you should insure that the executable name ends with the .exe extension or modify the gcc flags to include -Zexe.

It is likely that even with all of the changes and customizations you will still need to put some #ifdef statements in the source to get it to perform correctly for OS/2 use. I usually include a -DOS2 option in my makefile GCC compiler flags to permit a #ifdef OS2 line to be interpreted correctly in the source. Other choices include adding a #define line in config.h or using the automatically defined __EMX__ variable set by EMX itself. I find the OS2 variable to be the most intuitive when reading the code later. Once you have everything done why not share your results with others. Often the original author is more than happy to incorporate your changes into the original source and with your help, and perhaps a post to Hobbes' incoming directory, another great OS/2 program will be available for everyone to use. it is then displayed on the output window.