Re: TOOLS_PLATFORM.bash required version
Greg Troxel <[email protected]> Tue, 30 Jun 2026 18:18:44 -0400
| Newsgroups | gmane.os.netbsd.devel.packages |
|---|---|
| Message-ID | <[email protected]> |
Edgar Fu=C3=9F <[email protected]> writes: >> We do not seem to document what USE_TOOLS+=3Dbash means in terms of whet= her=20 >> the package can then expect bash 4, bash 5, or some particular 5.x. > How common is it for a package to need bash>3? > Would it make sense to have a "bash" tool plus a "bash4" tool? bash 4 was released over a decade ago. Nobody, except apple, thinks that using bash 3 is a reasonable thing to do. We do have two examples of programs that fail with bash 3. They cost a bunch of human hours to diagnose and work around. (I don't know how many.) I don't think it makes sense to add complexity and have multiple flavors for varying degrees of obsolete. The point is to minimize total human work. Programs that need bash usually don't mention it, and they ~never spec which versions are needed. I have proposed that TOOLS_PLATFORM.bash must be >=3D 5. That means systems with bash 4 can't set it, and bash will be built from pkgsrc on those platforms. A little cpu time, and taking lots of potential trouble off the table. So: Do you know of any actual systems that have a builtin bash that is 4? (Yesterday, macOS had builtin bash 3, and today it has no builtin bash.) Are these systems even remotely supported (security fixes from their upstream)? Do you see any real problem with making bash4-builtin systems build bash 5? Is there a plausible story that even one person will need to build bash 5, and not have to build cmake or something else that is substantially more bloated than bash? Alternatively, have you done a bulk build on such a system, and run it as a desktop, so there's a good basis to believe bash4 is ok for the many hundreds of packages that use it as a tool?