On 5/3/20 3:39 PM, Bruno Haible wrote: > Is it more important to be resilient, or more important > that if help2man fails for reasons that are in the package itself, "make" > fails?
We've been doing the latter for quite some time. I'm not sure the Haiku tar/kernel bug is a good argument to change common practice, as one can run into similar problems on more-common platforms if the timestamps on one's files are messed up. For example, on Fedora 31: $ echo 'a:b; cp b a' >Makefile $ touch -d@-1 b $ make make: b: Timestamp out of range; substituting 2514-05-29 17:53:03.999999999 make: Warning: File 'b' has modification time 15591323089 s in the future cp b a make: warning: Clock skew detected. Your build may be incomplete. $ make make: b: Timestamp out of range; substituting 2514-05-29 17:53:03.999999999 make: Warning: File 'b' has modification time 15591323087 s in the future cp b a make: warning: Clock skew detected. Your build may be incomplete. At some point it's probably better just to say "Don't do that".... PS. This example exploits the fact that GNU Make as currently configured will stop working on 2514-05-30 01:53:04 UTC. Presumably someone will need to fix Make before then....
