On 5/20/22 09:05, Rob Landley wrote: > And again, why a --show-stored-names if it's the default? There isn't an "rm > -r > --do-not-recurse". How does this HELP?
And of course, using that one/two/three/four/five/six hierarchy where six is a file and the ones before it are directories: diff -u <(tar c one --xform=s@three/four/@zero@ | hd) \ <(./tar c one --xform=s@three/four/@zero@ | hd) --- /dev/fd/63 2022-05-20 20:54:05.239891396 -0500 +++ /dev/fd/62 2022-05-20 20:54:05.239891396 -0500 @@ -40,14 +40,13 @@ 00000520 00 00 00 00 00 00 00 00 00 6c 61 6e 64 6c 65 79 |.........landley| 00000530 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................| * -00000600 6f 6e 65 2f 74 77 6f 2f 74 68 72 65 65 2f 66 6f |one/two/three/fo| -00000610 75 72 2f 00 00 00 00 00 00 00 00 00 00 00 00 00 |ur/.............| -00000620 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................| +00000600 6f 6e 65 2f 74 77 6f 2f 7a 65 72 6f 00 00 00 00 |one/two/zero....| +00000610 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................| * Because the filename the gnu/dammit version is actually SAVING INTO THE TAR ends with a / (it adds one to all directory entries!) but the slash was added AFTER it did the regex. Full bug-for-bug compatibility with this feature is just ACTIVELY STUPID. I think I'm just going to implement something that makes sense and wait for people to complain with specific test cases where they explicitly want the stupid. (Which means "how do I write test cases that both the host and toybox pass" remains annoying. My tendency to probe all the weird corner cases of a command to see what it does is not finding a coherent design here, it's finding bad implementation details bubbling to the surface...) Rob _______________________________________________ Toybox mailing list Toybox@lists.landley.net http://lists.landley.net/listinfo.cgi/toybox-landley.net