Portage tree have Erics Wassenaar's net-misc/host version (released on 1999). Now this package is maintained by Greg A. Woods who released the last version on August of this year. Also the current version in Portage Tree (991529) do not compile due this: emake CC="${CC}" COPTS="${CFLAGS}" || \ die "emake failed" There is no $CC defined on the enviroment, so the ebuilds fails...
Created attachment 42082 [details] host-20040812.ebuild
Created attachment 42083 [details, diff] files/host-20040812-scripts.patch
Created attachment 42482 [details] host-20040812.ebuild Anyway please check the obsolete ebuild, host provides buggy bash scripts (nslookup, et al.) they need to be fixed, meanwhile a host ebuild that only provides host. In my free time I will fix them, until that files/host-20040812-scripts.patch are not needed and IMHO the ebuild is ready for a version bump. Note: The old host (its already on the portage tree) do not provide that scripts.
Created attachment 48786 [details] host-20040812.ebuild
I like the Bug 80805 idea of change the category for this packages to net-dns.
Hi Octavio Thanks for reporting the update and sorry for long silence, - I was on a prolonged "away" and seems like nobody else cared enough.. Anyway, I am taking at this right now, and there are couple of questions: 1. what was the reason for specifying RESTRICT="nomirror"? I cannot see any need to restrict the mirroring. Looks like package is intended to be distributed under as-is or some such license, however do you know exactly which one? I'll email Greg Woods I guess, to clarify. 2. Another thing. I see you rename the binary/man page to host-woods. This is a reasonable thing to do to avoid overwriting host binary from the dns-tools package, but this will require modifications to xtraceroute strong/openswan and another one that make use of this utility. They naturally expect to find host, from whatever package you chose, in the path and use it and it is better to avoid modifying these packages. Instead we should think of whether to create a virtual/host and provide some host-config tool or choose some other way of dealing with the issue. Although I agree that it would be nice to take care of avoiding overwriting the binaries at this time, before we push an update in. I'll discuss this on gentoo-dev after the other two issues are resolved. 3. The package builds fine, but it doesn't do anything. Well, at least not anything it is supposed to do: aldar host # host-woods slashdot.org *** slashdot.org does not exist (Authoritative answer) while regular host finds slashdot fine. tcpdump shows an outgoing requests although (will check full traffic later, too busy network here), so I am not too sure what is the problem. George
Created attachment 63471 [details] somewhat cleaned version Hm, it may be that code is interacting with libresolv differently in this version. How did the produced binary behave on your system? Also, what ARCH are you on? I am testing this on amd64, so this might be the case.. Anyway, I am attaching a somewhat revampled ebuild - added RESTRICT="maketest" and moved two last sed commands to a commandline option of emake. Plus few cosmetic changes. (And this behavior happens no matter what changes I do - I tested it in and out just now). Lets see what we can do about it. George
(In reply to comment #6) > Hi Octavio Hi George. > Thanks for reporting the update and sorry for long silence, - I was on a > prolonged "away" and seems like nobody else cared enough.. np. :) > 1. what was the reason for specifying RESTRICT="nomirror"? Oh, I just forgot remove it. Its because I'm using host in my portage overlay and if RESTRICT="nomirror" is not set it will try to grab the source from gentoo mirrors (and there are no host-2004 sources) before grab it from the original source. > 2. Another thing. I see you rename the binary/man page to host-woods. This is > reasonable thing to do to avoid overwriting host binary from the dns-tools > package, but this will require modifications to xtraceroute strong/openswan > Instead we should think of whether to create a > virtual/host and provide some host-config tool or choose some other way of > dealing with the issue. I like that, but some tools doesnt work with dnstools host and other doesnt work with the Wood's one. So, Im still believing that renaming host to host-something (I just didnt like it the host-x) its fine. There are different tools, with different options although they do something similar and have the same name > 3. The package builds fine, but it doesn't do anything. Well, at least not > anything it is supposed to do: > aldar host # host-woods slashdot.org > *** slashdot.org does not exist (Authoritative answer) > while regular host finds slashdot fine. tcpdump shows an outgoing requests > although (will check full traffic later, too busy network here), so I am not > too sure what is the problem. Ugg, I tested that 9 moths ago, I'm sorry, I can not remember if its bad or if it just need some command line option in order to work, let me do some test in order to see what's appening. Thanks!
Hi Octavio. Any news with testing? I am still having the same problem.. George
Me too. :( I will contact Greg in this week. Did you do it before?
I did, but only regarding the licensing, at the very beginning. Here is his reply. [ On Friday, July 15, 2005 at 15:30:17 (+0200), George Shapovalov wrote: ] > Subject: "host" licensing > > First let me thank you for keeping an eye on a quality tool! Nice to see some > old, but usefull stuff kept in a proper hands. You're very welcome! I couldn't do without it myself! > I am a developer for a Gentoo Linux and I am about to process an update for > this package. Unfortunately I could not find any indications of what license > this version is released under, neither on the web site nor in the package > itself. See the comment at the top of each source file. If it's got a copyright then the copyright license is there. As for the rest of the files, and changes made since those licenses were dated, well I obviously I can't speak for Eric's estate. However for my own work I've always assumed that my changes are donated to the original copyright owner, under the original license, baring any separate notice.
Current version is 20070128; assigning this to maintainer-needed as it hasn't gone anywhere since 2004.
FWIW, the ebuild from comment #7 works fine here on x86 (version 20040812). And the exact same ebuild seems to work well also for version 20070128. I would like to suggest to rename this package from net-dns/host to something like net-dns/host-woods. I suspect the current name to be misleading (it was at least for someone else and me), making people believe they will get the usual 'host' binary if they install this package, while that one is actually provided by net-dns/bind-tools. (And while at it, maybe the man page should be adjusted to say 'host-woods' instead of 'host'.) Finally, I couldn't find any documentation about RESTRICT=maketest used in this ebuild. What does it mean? Same as RESTRICT=test?
(In reply to comment #13) > Finally, I couldn't find any documentation about RESTRICT=maketest used in this > ebuild. What does it mean? Same as RESTRICT=test? It's invalid.
Ah yes, this one. Just wanted to mention that I tried to revisit it "recently" (a month or two ago) and apparently the package can be linked vs different resolver libs, but selecting the right one was not quite straightforward (IIRC this is done via some config file. So, potentially, detecting, based on running arch, the right one might be necessary and then seding that config file). Sadly I did not have much time for this package recently. However whoever tries to process it should keep in mind that trouble does not end with the resolver library. One should keep in mind that this thing would clash with host from bind-tools, as was mentioned above. Nonetheless simply renaming the thing is not optimal, as some packages might want to use it preferentially over bind's host. Therefore some handling of this, probably via some eselect module would be necessary (akin to what was done for ctags recently for example). This version of host provides a superset of functionality from what I understand, so, theoretically, it can be a drop-in replacement for bind's host. This is according to the description, I haven't tested this personally, as it ceased to work for me few years ago. But this makes the situation with this and bind's host(s) (again, theoretically) suitable to be resolved via eselect module and, probably, appropriate virtual. George
Created attachment 129405 [details] host-20070128.ebuild w/o static linking to libc, other fixes The previous ebuild attachment built statically, which included libc. Currently, I still have it linking to libresolv.a, as trying to use just -lresolv results in this: file.o: In function `cache_write': file.c:(.text+0x140): undefined reference to `__ns_put16' send.o: In function `host_res_write': send.c:(.text+0x3a9): undefined reference to `__ns_put16' collect2: ld returned 1 exit status The symbol is indeed not in libresolv.so; I don't know why it's not. I haven't looked into it much. Other fixes: One of the typedefs used was broken by not checking for x86_64, which basically ended up causing any IP address entered to trigger an error when built with any -Olevel. Speaking of -Olevel, the Makefile was defining it's own, twice; -O2 then -O, with -O being the last and therefore overriding any user CFLAGS. Lastly, it was also setting -pipe; I figure that's also up to the user, so it went as well.
Thanks Thomas! That fixed the linking issue. Seems to work fine here, on amd64. I added this version to the tree. As the only app that used the hostx form - the xtraceroute, has been pulled from the tree I retained the host-woods name. We can go about creating an eselect module if there is an interest expressed. May make sense to re-add xtraceroute back first as well (if I am not mistaken, this package was a major reason why it was pulled). For now I'll close the bug. If people want xtraceroute back and an eselect module, please reopen and add comments correspondingly.. George
