To install buildsome follow these instructions.
Tired of cryptic bugs in the build system? Of ever having to run "make clean"? Tired of manually specifying dependencies and introducing ad-hoc dependency scanners? Then buildsome is for you!
Buildsome is an innovative build system, meant to both ease the declaration of the build steps, and give better guarantees to users.
As long as the build steps being used depend only on inputs from the file system, buildsome guarantees that the build is correct and determinstic.
The dependencies/inputs are automatically detected.
With buildsome, you're free to specify your inputs partially (such specification can aid a more efficient build process). You do not have to specify your file system inputs explicitly though, as they're automatically detected using file system access hooks.
Automatic detection of dependencies does not only add convenience, it also guarantees correctness. Correctness means no "false negative" builds. "False negative" means that when inputs change, the build system does not always recognize this properly, and avoids rebuilding parts that require it. Such false negatives plague virtually every build system in existence.
As an example: Virtually all build systems use gcc -M to detect
#include dependencies. If gcc was invoked with -Ia -Ib, and
example.h was included and found in b/example.h, most build
systems will only know to rebuild if b/example.h is changed. If
a/example.h is introduced, however, the compilation process will
have a different result, and buildsome will detect this and re-execute
the compilation command, unlike other build systems.
Another correctness issue solved by buildsome is meddling with the file system during a build process. Unlike other build systems, buildsome detects this and will not record false build results in detectable cases. The only undetectable cases are third-party meddling with outputs of a command while it is being built. Editing a source file immediately after it was fed to a compiler will not accidentally mark the compilation as "newer" than that source file.
Virtually all build systems do not capture stdout/stderr of executed commands. When a command succeeds (e.g: a compiler invocation), its stdout and stderr are displayed once and then forgotten. All future builds will never present the warnings from the compiler, which are thus lost until a "clean" command is issued. Buildsome captures these outputs and re-prints them when an execution is cached. This avoids the most common reason "-Werror" is used: as warnings are no longer lost.
Automatic detection of file system dependencies and parallelism are somewhat contradictory. To overcome this, buildsome uses speculative parallelism based on previous detected inputs. These are likely to be the inputs of the new executions as well, so are speculatively built.
The global file system hooks can facilitate a generalized "ccache", that works correctly with more than just a C compiler. It is planned to support caching of all build results, rather than just the last one. Switching between debug and release builds should not require rebuilding everything, if such versions were built in the past. Additionally, "global" caching should be available: On a company LAN with many builders, it is often more efficient to pull build results from the network than to rebuild them locally.
Targets should be more like functions that can be parameterized. File names will have an auto-generated hash component in them to represent the parameters used. This will allow, for example, painlessly building the same libraries in various different configurations to link into different executables without duplicating the libraries' declarations.
When speculative parallelism is detected as unnecessary, buildsome will terminate the speculations.
Use "cabal install" to build and install buildsome.
The file system hooks are inspired by tup. They are taken a couple of steps further though, by hooking the file system globally, and not just locally, avoiding false negatives relating to system-wide changes, too. These hooks are also used to auto-detect dependencies and delay dependent executions, rather than just verify explicitly specified dependencies.