From Fedora Project Wiki

Fedora Classroom - Package Taxonomy and Techniques - Ignacio Vazquez-Abrams - Saturday, November 8, 2008

IRC Log of the Class

-!- nirik changed the topic of #fedora-classroom to: Fedora Classroom - Package Taxonomy and Techniques with your teacher: ivazquez - See Communicate/IRC/Classroom for more info 23:00
Ineluctable thanks stickster 23:00
stickster neverho0d: It is, but you could work with another person to edit your summaries 23:00
* kdn *bell rings* 23:00
brunowolff If it isn't, then translating it could be another area to help out. 23:00
neverho0d stickster: thank you 23:01
stickster brunowolff: +1, exactly 23:01
* stickster really scoots now 23:01
ivazquez Hello everyone, and thank you for coming. 23:01
mattia hi ivazquez 23:01
Abd4llA Helloo Mr ivazquez :P 23:01
neverho0d brunowolff: I'll be glad to contribute as much as I can 23:01
thomasj helloiva 23:01
thomasj ops 23:01
ivazquez You see them go by all the time in anaconda and PackageKit. 23:01
kiakli hi 23:01
cga ciao ivazquez, i'm here too ;) 23:02
thomasj hello ivazquez 23:02
ivazquez libfoo-3.2-5.fc9... barprogs-5.2.6-0.cvs20081016... 23:02
ivazquez But just what is it that makes a package... 23:02
ivazquez ... a package? 23:02
domg472 rpmbuil 23:02
domg472 nvm 23:02
brunowolff A spec file. 23:03
ivazquez At its simplest, a package is a set of files, and metadata about each file as well as about the entire bundle. 23:03
kdn clean installs, clean uninstalls, clean upgrades? 23:03
ivazquez Things like permissions, ownership, scripts executed during installation, etc. 23:03
ivazquez So let's spend some time looking at these things. 23:04
ivazquez The most important thing you need to know about a package is its NEVRA. 23:04
ivazquez This stands for Name, Epoch, Version, Release, and Architecture. 23:04
ivazquez It's what determines when a package can be installed or upgraded. 23:04
ivazquez The name is the part we're most familiar with. 23:05
ivazquez glibc, firefox, etc. 23:05
ivazquez It is the smallest part of a package we can use and still talk about the package itself. 23:05
ivazquez The epoch, version, and release determine the "order" of packages. 23:06
ivazquez That is to say, which package is considered "newer" than others. 23:06
ivazquez Normally we don't need to worry about the epoch, so we'll hold out on that one for just a moment. 23:07
ivazquez Version is mostly self-explanatory. 23:07
ivazquez 2.8, 3.0.2, etc. 23:07
ivazquez When upstream releases software, they give it a version to keep it separate from all the other versions they release. 23:07
ivazquez The first part of a package that we really get to handle as a distro is the release. 23:08
EvilBob how should a packager deal with packaging a beta or prerelease version from upstream? 23:08
ivazquez That is actually covered in Fedora's packaging guidelines. 23:09
ivazquez I'll be happy to discuss packaging issues after. 23:09
@nirik 23:09
ivazquez The release is used to distinguish one package of a specific version from another. 23:09
ivazquez So you package libfoo 3.2, and the first package gets a release of "1". 23:10
ivazquez Then let's say someone notices a problem in your package, and you fix it. 23:10
ivazquez When you go to fix it, you keep the version the same (since it's still the same code from upstream), but you bump the release up to "2". 23:11
ivazquez That way people and tools realize that libfoo-3.2-2 is an update for libfoo-3.2-1. 23:11
ivazquez And when upstream releases a new version, say, libfoo 3.3, you can drop the release back down to "1". 23:12
ivazquez libfoo-3.3-1 is seen as an update to libfoo-3.2-2 since the version is higher. 23:12
brunowolff The kernel seems to keep an increase release number even after minor releases? 23:13
ivazquez It used to. They fixed that. 23:13
ivazquez The exact details involve all sorts of ugliness regarding the "base" commit. 23:13
ivazquez We can come back to that later. 23:14
brunowolff When 2.27.5 was released it had -88 for the release number? 23:14
ivazquez Now, epoch. 23:14
brunowolff OK 23:14
ivazquez The epoch is the rarest-seen and the most troublesome part of a package. 23:14
ivazquez By default it's not shown when you use rpm to query a package. 23:15
ivazquez (and we'll come back to querying after) 23:15
zcat if you have rpmdevtools installed you can play with the version/epoch comparisons. rpmdev-vercmp libfoo-3.2-2 libfoo-3.2-1 23:15
ivazquez Additionally, when writing a spec file, you must include this invisible epoch in listing packages that you are dependent upon. 23:16
ivazquez Many a packager has been bitten by omitting the epoch in this scenario. 23:16
ivazquez Now, why would anyone use the epoch, given all this trouble? 23:16
ivazquez The simple answer is that the version has had issues. 23:17
ivazquez For instance, going from, say, "3.2preview" to "3.2final" will be a problem, since 3.2preview actually looks newer than 3.2final. 23:18
cga how comes? 23:19
ivazquez So what must be done is the epoch must be bumped to 1 (it is 0 by default), and everyone involved lets out a collective groan because of the issues mentioned earlier. 23:19
thomasj cga, p - f 23:19
Sid because 2.3preview comes later in an alphabetical listing than 2.3final 23:19
ivazquez Strings in a VR are compared lexicographically. 23:19
Sid 3.2 even 23:19
cga i see thanks 23:19
EvilBob cga: alphanumeric ordering 23:19
cga fair enough 23:19
ivazquez This brings us to the fifth part, the architecture. 23:20
ivazquez You would think that this is a very small topic. 23:20
ivazquez After all, i386 goes on i386 machines, x86_64 on x86_64, and so on. 23:20
ivazquez Unfortunately this is not true. 23:21
ivazquez And this is for 2 reasons. 23:21
ivazquez The first is that there are architectures that are "equivalent" to each other. 23:21
ivazquez E.g., i386, i486, i586, i686, etc. 23:22
ivazquez You can only have a single package (with exceptions we'll cover later) with the same NA installed at a time. 23:22
ivazquez So you can't have libfoo.i386 installed at the same time as libfoo.i686. 23:23
ivazquez The second is a little feature called "multilib". 23:23
ivazquez This allows packages from different equivalent architectures to be installed at the same time. 23:24
ivazquez It's what lets you install glibc.x86_64 and glibc.i686 on the same system. 23:24
ivazquez However, it also has its issues. 23:25
Discordian Doesn't that increase the memory residency? 23:25
ivazquez One of the issues is when the x86_64 package and the ix86 (i386, i686, etc.) package contain the same files. 23:26
ivazquez Sorry, memory residency? 23:26
Discordian Don't shared libs take up memory? 23:26
Discordian Sorry ignore that please don't want to distract you 23:27
ivazquez Only when they're loaded. Packages on the disk take up no memory until they're requested. 23:27
ivazquez Alright, same files. 23:27
cga Discordian: see differences between application and process (ie. installed versus running) 23:28
ivazquez rpm does not have a problem with the x86_64 package and the ix86 package containing the same files, provided that the files are *exactly* the same. 23:28
ivazquez Timestamp, contents, permissions, etc. 23:28
Discordian cga: thank you 23:28
ivazquez But even this is not true. 23:28
ivazquez And this is where we get into trouble. 23:28
ivazquez If the x86_64 package and the ix86 package are installed in the same "transaction", rpm is perfectly willing to overlook differences in the files and will install both packages, with the x86_64 files overwriting the ix86 files. 23:29
ivazquez HOWEVER. 23:29
ivazquez If either one of the packages is already installed, and we try to install the other package, rpm will complain about file conflicts since they don't match. 23:30
ivazquez This sort of issue has spawned many more bug reports than are deserved. 23:31
ivazquez So, something to keep in mind when you install or build packages. 23:31
ivazquez So, now that we understand (I hope) these 5 elements, let's look at what else a package's metadata provides. 23:32
@nirik I have a question: where does rpm store the ix86 binaries? If the x86_64 package is removed it would have to place the ix86 ones back right? 23:33
ivazquez It does not. It simply doesn't erase the files provided by the x86_64 package. 23:33
ivazquez Yay multilib. 23:33
@nirik yikes. ;( 23:34
Discordian Does that mean you could rebuild both i386 and x64 rpms using rpmrebuild and they'd both be valid? 23:34
ivazquez I don't know. I've never looked at how rpmrebuild works. 23:35
Discordian My impression is it just uses the rpm db to reconstruct the rpm file 23:35
ivazquez I'm glad you bring up the rpmdb. 23:36
ivazquez A package in a file has a set of metadata. 23:36
ivazquez When said package is installed, the metadata needs to be stored in order to retain information about the package. 23:36
ivazquez The metadata is stored in what is called the rpmdb. 23:37
ivazquez We can query the rpmdb via "rpm -q". 23:37
ivazquez We pass it a name, and rpm will give us the metadata in a pre-programmed format. 23:38
ivazquez Normally it looks like <name>-<version>-<release>. 23:38
ivazquez I'm not sure if they've added .<arch> to that recently. I believe so though. 23:38
ivazquez You can see an example by opening up a terminal and typing "rpm -q filesystem". 23:39
Discordian That is the case on F9 yes 23:39
ivazquez If someone could provide the output from theirs please? 23:39
Sid filesystem-2.4.13-1.fc9.i386 23:40
ivazquez Right, there we are. 23:40
ivazquez The name is, of course, "filesystem". 23:40
ivazquez The version is "2.4.13", the release is "1.fc9", and the arch is "i386". 23:41
EvilBob [root@mediapc1 ~]# rpm -q filesystem 23:41
EvilBob filesystem-2.4.13-1.fc9.i386 23:41
EvilBob grrr 23:41
ivazquez Sometimes we need more or less than is provided by the default format. 23:41
zcat ah. so showing the arch is default now? i've been adding "%_query_all_fmt %%{name}-%%{version}-%%{release}.%%{arch}" to ~/.rpmmacros so i didn't notice 23:41
EvilBob filesystem-2.4.11-1.fc8 23:41
Discordian filesystem-2.4.13-1.fc9.i386 is what I have 23:41
ivazquez Right, and we're get to query formats right now. 23:42
ivazquez *getting 23:42
ivazquez Let's say that you have a script that verifies the version of certain packages on your system. 23:42
ivazquez You want to narrow down your focus to only the version. 23:42
ivazquez Normally you'd do post-processing of the output of "rpm -q". 23:43
ivazquez But the default output can be changed, as demonstrated by zcat just above. 23:43
Discordian you mean using awk or perl? 23:43
ivazquez Correct. 23:44
ivazquez What you can do is you can pass rpm a parameter that tells it the format you want. 23:44
Discordian cool I can see that would be useful 23:44
ivazquez This argument is "--qf", and it takes the format in a special, err, format. 23:44
ivazquez Let's deal with just the version for now. 23:45
ivazquez To get the version of filesystem we run "rpm -q --qf '%{version}'". 23:45
ivazquez This will return just the value, with no fancy formatting. 23:45
ivazquez Naturally if the output will be consumed by a person we'll at least want to put it on its own line. 23:46
ivazquez The query format string takes several C-style escapes, including \n for a newline. 23:46
nuonguy you mean rpm -q filesystem --qf '%{version}' right? 23:47
ivazquez "rpm -q --qf '%{version}\n'" will give us something that we can deal with easier. 23:47
ivazquez Yes. 23:47
neverho0d rpm -q --qf '%{version}\n' filesystem 23:47
ivazquez Right. 23:47
ivazquez There are a large number of these "query tags" built into rpm. 23:48
ivazquez You can see a full list by running "rpm --querytags | less". 23:48
ivazquez The query tags in the query format are not case-sensitive. 23:48
neverho0d wow 163! 23:48
ivazquez Most of them you'll never need though. 23:49
daMaestro and shortcuts for those that don't want to type ;-) 23:49
ivazquez Among the important ones are the 5 parts that we started with here: 23:49
ivazquez name, epoch, version, release, arch. 23:50
ivazquez Also important is sourcerpm, which gives you the name of the component in Bugzilla to file bugs against. 23:50
ivazquez installtime is also useful. 23:51
ivazquez But there are 2 things to know about it. 23:51
ivazquez 1) It only makes sense in the rpmdb. Querying it from a package in a file makes no sense. 23:51
ivazquez And we'll get to querying a package in a file in a moment. 23:51
ivazquez 2) It gives its result as a *nix timestamp. Not exactly human-friendly. 23:52
ivazquez So "rpm -q --qf '%{name}: %{installtime}\n' filesystem", while technically correct, isn't all that useful. 23:53
ivazquez Fortunately query tags support a suffix that you can add to modify them. 23:53
ivazquez The suffix ":date" will convert a timestamp into a human-readable date and time. 23:54
ivazquez So with a small change we get "rpm -q --qf '%{name}: %{installtime:date}\n' filesystem", and we can see more clearly when our system was installed or upgraded. 23:54
ivazquez Another query tag of interest is dsaheader. 23:56
ivazquez This tag tells us how, when, and who signed a package. 23:56
ivazquez We'll get into signing packages a bit later. 23:56
Discordian Cool, this is good stuff 23:57
cga night all, this is interesting but i'm falling asleep. i'll read the logs tomorrow. i find the classroom idea very interesting. kudos. 23:57
ivazquez Take care. 23:57
ivazquez Again, it spews a long hex string that will make our brains asplode if we try to consume it. 23:57
* nirik didn't know the :date stuff. very cool. 23:57
ivazquez Suffixing it with ":pgpsig" processes it and gives us a result we can use. 23:58
daMaestro yes, i just wasted a few moments trying to pipe the unix timestamp to data -d via xargs 23:58
daMaestro thanks. 23:58
kdn Thanks to all. This was a Good Thing(tm). 23:58
* delhage agrees 23:58
ivazquez rpm -q --qf '%{name}: %{dsaheader:pgpsig}\n' filesystem 23:58
daMaestro it's not over 23:58
Discordian I'm not going 23:58
ivazquez I'll leave it as an exercise for the reader as to whose key that is ;) 23:58
Discordian heh 23:59
brunowolff How well do you have to know the code to package something from upstream? 23:59
* nirik notes that this is the last class scheduled today... more classess tomorrow. 23:59
ivazquez Almost not at all. 23:59
domg472 why --qf and not -qf, would that conflict? 23:59
brunowolff So a good working relationship with upstream would be good enough? 00:00
ivazquez -qf is seen as a combination of 2 flags, -q and -f. 00:00
domg472 i see 00:00
Discordian -qf is which package does this file come from 00:00
ivazquez brunowolff: Not even. There are a number of instances where upstream is not even aware of it. 00:00
ivazquez Okay, packages on disk. 00:00
Sid is it possible to query packages that aren't installed? 00:00
ivazquez Yes it is. 00:01
ivazquez There isn't much to say about packages on disk, other than you need to add -p to rpm. 00:01
ivazquez And of course, some tags such as installtime don't make sense. 00:01
brunowolff Is there a good place to get info on how to package java apps? I have one I build from ant locally, but I don't know where fedora wants 00:01
ivazquez After all, a package on disk isn't considered to be installed. 00:02
brunowolff java apps placed or if there are any particular build practices to use. 00:02
neverho0d ivazquez: it would be helpful to explain a package breaking on docs, devel and other things.... 00:02
Sid ok, what if the package isn't on disk but simply in the yum repo, can I then query it? 00:02
ivazquez brunowolff: Fedora has an extensive set of packaging guidelines in the wiki. 00:02
Sid or do I at least need to download the .rpm? 00:02
ivazquez Another excellent question. 00:02
ivazquez neverho0d: We'll handle that in a bit. 00:02
ivazquez Packages in a repo. 00:03
neverho0d ivazquez: ok. thnks 00:03
ivazquez yum-utils has a script, repoquery, which can be used to query packages in a repo. 00:03
@nirik brunowolff: Packaging/Java 00:03
ivazquez It takes a subset of the query tags that rpm does, due to the fact that the repo metadata has only a small fraction of the package metadata. 00:04
ivazquez Other than that it takes -q and --qf just like rpm. 00:04
ivazquez Alright, I think I've chewed up enough time on my bit. Which topics did I say I'd defer? 00:05
brunowolff nirik: Thanks that looks pretty readable. I'll see if I can make it work locally. There are other issues with packaging the app (Colossus) that I am interested in. 00:05
* nirik is sad that it doesn't provide sourcerpm. ;( 00:05
domg472 thanks 00:05
domg472 very informative day 00:05
ivazquez Package breakup. 00:06
Discordian Thank you very much 00:06
erinlea80 Thanks ivazquez :) Its been informative. 00:06
Discordian I learned a lot 00:06
ivazquez Not quite done yet, but I understand if you have other things to do. 00:06
mattia thanks ivazquez 00:06
Discordian No no I'll stay 00:06
Bugz thanks ivazquez 00:06
ivazquez Part of the metadata in a package is what "type" a file is. 00:06
ivazquez That is to say, whether it's a normal file, a configuration file, or documentation. 00:07
Discordian or a script? 00:08
ivazquez rpm allows you to query only these specific files by passing -c or -d, but most of the query tags are at the package level and so don't apply. 00:08
ivazquez Package scripts are actually stored separately. 00:08
ivazquez They're in the metadata, not the files. 00:08
Discordian Okay I've wondered about that 00:08
ivazquez They're viewable as either any of the *prog query tags, or by passing --scripts when querying a package. 00:09
domg472 rpm -q --scripts 00:09
Discordian Thank you 00:09
ivazquez No, I lie. Not the *prog tags. 00:09
ivazquez Those are what is used to actually run the scripts. 00:10
ivazquez The scripts themselves are in the prein, postin, preun, postun, and other related tags. 00:10
ivazquez Okay, I can't seem to find what other audience topics I decided to defer. 00:11
ivazquez Does anyone have any questions?] 00:11
domg472 it was all clear except the x86 x86_64 bit but that must be my fault 00:12
nuonguy is there a virtual file system like /selinux or /proc to browse the rpm database? 00:12
domg472 ill digginto that myself 00:12
Discordian since this is Fedora : what impact will the new version of rpm have? 00:12
nuonguy domg472: +1 00:12
Discordian or is that unfair? 00:12
ivazquez No. The rpmdb is stored as a set of bdb or sqlite databases in /var/lib/rpm. 00:12
erinlea80 My questions were answered up above. 00:13
ivazquez Which one, rpm5? 00:13
* nirik muses that someone could make a fuse-rpmdb someday... 00:13
Discordian Yeah 00:13
domg472 by the way nuonguy id like to hear what is your reason to contribute to fedora or oss in general? 00:13
ivazquez rpm5 is an independent project. I cannot speak for Red Hat, but it has little or no impact in Fedora as far as I can see. 00:14
ivazquez the rpm maintainers in RH could probably give you a more complete answer on that one. 00:14
Discordian Okay I noted that F10 had flagged up a new version of rpm 00:15
Discordian I think perhaps it's 4.x 00:15
ivazquez That would be further development in the 4.x branch. 00:15
Discordian Yes 00:15
@nirik yes, 4.6.0 00:15
sdodson F10 will use 4.6. 00:15
sdodson Though, I believe it's been requested that newer features of 4.6 not actually be used until a later date. 00:15
Discordian Ahhh cool 00:16
ivazquez Anyone else? 00:16
@nirik thanks ivazquez. Great session! 00:16
ivazquez In that case, thank you for coming, and thank you for listening. 00:16
Discordian yes thank you very much ivazquez 00:17
nuonguy domg472: when I interview candidates, I ask them if they contribute to any open source projects 00:17
* erinlea80 applauds 00:17
neverho0d ivazquez: may be one another class om spec-file building? ;) 00:17

Generated by 2.7 by Marius Gedminas - find it at!