We have three awk implementations in gx86: gawk, mawk and busybox Some scientific package would profit from that. We would need: -virtual/awk -app-admin/eselect-awk I have added everything for testing in cj-overlay.
busybox doesn't provide an awk symlink, and our system requires gawk. so what exactly would virtual/awk accomplish when gawk is always going to be installed ?
(In reply to comment #1) > busybox doesn't provide an awk symlink, and our system requires gawk. symlink or not, "busybox awk" is an awk implementation. I have tested it with some of my science scripts. > so what exactly would virtual/awk accomplish when gawk is always going > to be installed ? mawk is significantly faster than gawk. busybox would save you the install of gawk of small systems. virtual/awk would ensure that at least one of them is installed. The 3 calls in misc-functions.sh of portage should work with any awk interpreter. In the eclasses mostly '{print $X}' is used, works with any awk interpreter.
awk is used in a lot of places, not just portage. just grep for it in /usr/bin. further, changing `awk` from gawk without consulting gentoo-dev is a no go. it is used heavily in the tree, and is what everyone develops with. so changing it implies a new support burden that everyone has to buy into. the fact that it's faster really doesn't matter for most of our things as it rarely gets called.
(In reply to comment #3) > awk is used in a lot of places, not just portage. just grep for it in > /usr/bin. Yes, but I haven't had any problems with mawk as interpreter so far. And as long as no secret GNU extensions are used, mawk will understand most of the GNU thing as well. For nawk, this is a different question. > further, changing `awk` from gawk without consulting gentoo-dev is a no go. > it is used heavily in the tree, and is what everyone develops with. so > changing it implies a new support burden that everyone has to buy into. True, this bug is preliminary work and thought collection prior a RFC on gentoo-dev. > > the fact that it's faster really doesn't matter for most of our things as it > rarely gets called. We could also add a mawk use flag to the sci package, which make extensive use of awk, and would profit from its speedup. But the solution don't seem very nice to me. Actually I came across this idea as Ubuntu ships mawk instead of gawk and I had to fix some awk code in one of my packages.
(In reply to comment #4) mawk supports POSIX, not GNU extensions. thus it isn't hard to locate the stuff that mawk doesn't support but is used. but that isn't the only lurking issue that causes support problems ... there are known edge cases where mawk works one way and gawk works another. like substr() and an index of 0. it can probably be made to work by requiring gawk and encouraging people to use `gawk` in the ebuilds, and to only use `awk` if they actually care about it. in the default install, only gawk would be installed (since nothing would pull in mawk). this would allow mawk to be installed and manage `awk` w/out affecting the ebuild environment.
(In reply to comment #5) > mawk supports POSIX, not GNU extensions. thus it isn't hard to locate the > stuff that mawk doesn't support but is used. but that isn't the only > lurking issue that causes support problems ... there are known edge cases > where mawk works one way and gawk works another. like substr() and an index > of 0. mawk know slightly more that POSIX, but this is all in the undocumented gray zone. > it can probably be made to work by requiring gawk and encouraging people to > use `gawk` in the ebuilds, and to only use `awk` if they actually care about > it. in the default install, only gawk would be installed (since nothing > would pull in mawk). this would allow mawk to be installed and manage `awk` > w/out affecting the ebuild environment. Sounds good, CCing sci to hear some more opinions.
Last time I checked, sys-libs/glibc fails to build unless `awk` points to `gawk`. Glibc explicitly runs `awk` during its build and requires that certain GNU extensions not provided by busybox or mawk be present in awk. Also (at least at that time), glibc upstream refused patches to call `gawk` instead.
(In reply to comment #7) i'll sort glibc out if that's still true
To avoid flooding with comments of incompatibility of !gawk, I created bug #418473 to keep track of packages broken by the usage of virtual/awk instead of sys-apps/gawk.
