The project borland-reborn has successfully reconstructed the source code for the Borland C++ 2.0 compiler, originally released in 1991. By using the original toolchain, the reconstructed source code compiles into a BCC.EXE binary that is byte-for-byte identical to the official version.
The codebase consists of 298 files across 15 subsystems, including 265 C files and 33 manually written TASM modules. The restoration process involved identifying approximately 3100 functions, global variables, and labels, along with 300 record fields and 40 record layouts, all documented with around 10 000 lines of comments.
To ensure accuracy, the team used the Exhume toolkit to verify every function and relocation. The build process requires original Borland distributions, specifically Borland C++ 2.0 and Turbo C 2.01, which are linked using TLINK 4.0 with the VROOMM overlay manager. The final output is validated via SHA-256 checksums.
The research uncovered several technical quirks in the original compiler, such as a manual recursive descent parser and specific bugs in library functions like strrchr and strcmp. Additionally, the team identified instances of dead code and limitations in bit-field handling that cause compilation failures in certain memory models.
The project is hosted on GitHub and distributed under the MIT license for the scripts and documentation. While the source code is provided for study and preservation, the project makes no claims to the copyright of the original Borland software, which remains the property of its respective owners.
