On Sun, Apr 23, 2017 at 01:04:56PM +0200, Holger Freyther wrote: I like the central handle on dependencies, but I don't really like XML. Why not use the git submodules approach? Can't submodules tag specific revisions? ~N

Hi Holger, On Sun, Apr 23, 2017 at 01:04:56PM +0200, Holger Freyther wrote: I would still suggest that the "release" tags should also be made in all the individual repositories. Otherwise, while looking at git history in an individual repository during development/review, it is not obvious which patches are in which release or not. So having the specific 'git commit' IDs into repo and bypassing release tags inside the individual git repositories, you are taking this information out of the individual repositories. I don't generally think that's a good idea. So if at all, I believe 'repo' should refer to tags inside the inidvidual repositories [not sure at possible], or the repository-individual tags should be auto-generated in the same process when generating a new xml file in repo. -- - Harald Welte <laforge(a)gnumonks.org> http://laforge.gnumonks.org/ ============================================================================ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6)

I would like to be lazy but if the have pending SO_VERSION bumps we need to make a proper release (bump the version, update the debian package). Maybe to make it a bit more easy drop a version scheme like v1.2.3 and have a strictly monotonic increasing number like used for systemd/udev (and then still have v123.X for minor releases) or the date approach I mentioned before. If we go for date as version number we should be able to automate most of it and assist where manual work is needed (e.g. work on the TODO-RELEASE). holger

Let me comment as well as recap the things discussed on OsmoDevCon: On Mon, Apr 24, 2017 at 01:26:20PM +0200, Harald Welte wrote: At OsmoDevCon as one of the very last items, we basically agreed for year+month style version numbering, and that we should not use just "Osmocom" for it, since Osmocom is much more than our core network + BSC (etc) infrastructure. The name floating around so far has been "Osmocom-Cellular". This makes a name and version scheme of e.g.: Osmocom-Cellular 17.04 Just now I notice that this would actually blur/confuse with the name of Open Cellular at least when abbreviated. Or maybe that's a good thing? Also Osmocom-BB is technically also part of "Cellular" but will not be part of what we plan to wrap in a release. (The plan is not to include *all* of the osmocom repositories in a release, right?) It occured to me that "osmo-net" or "osmo-network" or "OsmocomNetwork" (explicitly without the 'core') could be a good name for releases. What do you think about that? For example, we can call the releases: "osmocom-net-17.04" "osmocom-net-17.07" etc. In a separate discussion, we've also been looking for a name to replace "openbsc.git", to combine the MSC+BSC+SGSN+MGCPGW realms. The conclusion of *that* name discussion so far has been that we will instead try to split up into many smaller repositories: one for the MSC, one for the BSC, one for the SGSN... Does the Osmocom-Network name fit this? We should not use the name for both: the release also includes libosmocore, libosmo-*, osmo-bts, osmo-iuh, osmo-hlr, while the repository would have only been for MSC+BSC+SGSN+MGCPGW. So I would use the "Network" name for the release, and I would still go for separation of openbsc.git into smaller entities. To be contained in a release, RFC: libosmocore/ libosmo-abis/ libosmo-netif/ libosmo-sccp/ libsmpp34/ openggsn/ openbsc/ --> split as osmo-msc/ osmo-sgsn/? osmo-ps/? osmo-bsc/ osmo-hlr/ osmo-iuh/ osmo-bts/ osmo-pcu/ osmo-trx/ osmo-python-tests/ asn1c/ libasn1c/ sysmo-bts/layer1-api/ octphy-2g-headers/ osmo-gsm-manuals/ (Any additions or removals to this list?) What if a release turns out to contain a critical bug? The simplest way we would "resolve" that for now is to tell the users to revert to a release that works and wait for the next regular release. Nevertheless, it would be good to have a versioning scheme that allows for future changes in the release schedule without breaking the naming conventions. If we would like to tag several releases within a month, we could append a minor version like "osmocom-net-17.07.2". Should we add a ".0" by default? "osmocom-net-17.04.0"? Should we drop the leading zero from the month? "osmocom-net-17.4.0"? My personal choice for release names converges to: osmocom-net-17.6.0 osmocom-net-17.9.0 osmocom-net-17.12.0 osmocom-net-17.12.1 <-- critical bugfix release ("backed by funding") osmocom-net-18.3.0 Please comment on the naming -- the time window for permitted bike sheds is still open now. I suggest that we follow a regular release schedule of: at the beginning of March, June, September and December. (so that we don't need to roll a release during/right after the festive season, i.e. not on 1st January) More details on releases: We will so far provide simple tagging of an otherwise arbitrary state of the master branches -- the aim is merely to provide a common handle to users to be able to talk about the same "version" of our software, to allow using the same software state without the need to pass individual git hashes around. That means: - we will *not* maintain older releases with backports, - we will *not* do any special verifications of the working state of release X beyond the common sensical checks. - We *will* require a release to pass all stock 'make check' and 'make distcheck' (as we do for all our patches), - *and* we will require the osmo-gsm-tester setup to be in production before commencing releases (I'm on it), so that - we will require a release to pass the osmo-gsm-tester physical tests. If anyone is interested in maintenance of older releases or more extensive release management, you are a) free to do so and/or b) request such, backed by appropriate funding (for a longer term, probably). Our releases will be scheduled every three months for the time being. The individual git repositries will remain separate, and each will have their own version (maintained in signed git tags to be used for compatibility checks in autoconf). The release name should also be tagged in each of the individual git repositories (without interfering with 'git describe', i.e. non-signed non-annotated git tags, best prefixed by "release_" or "release/" to allow looking it up automatically?). The combined release repository (be it by repos or by git submodules) shall reference those release tags instead of individual commit hashes (basically to make sure we do have such tags in place). Upon invoking 'osmo-foo --version', we would ideally output both the individual 'git describe' version as well as the larger release version sort of like ▶ osmo-msc --version OsmoMSC version 0.15.0.735-1759 (osmocom-net-17.6.0) What if it's from a bleeding edge master, i.e. most-recent-release-tag != HEAD? ▶ osmo-msc --version OsmoMSC version 0.15.0.735-1759 (osmocom-net-17.6.0-dev) Or maybe rather a scheme combining both? like: OsmoMSC version 17.6.0-0.15.0.735-1759 Though the library version used for autoconf checks should not contain the release name -- so I would personally opt to keep these entirely separate, meaning that we would add a separate variable to the config.h (like $RELEASE) determined by a separate script. Loosely related and still open is the question of a scheme to effect version bumps of the individual projects. I had at some point started to do a tagged version bump for each API addition to e.g. libosmocore, but that was a lot of overhead (including adjusting configure.ac of dependent projects). It was somewhat agreed that we would collect significant API changes in the TODO_RELEASE text file and bump a version tag whenever a "new release (TM)" would be made ... but we never defined when we would do that nor how exactly. So, added bonus would be to resolve this question along with this discussion, alas it might blow up the bikeshed permutations... Naive question to clarify: if we have overall release tags, why exactly do we need individual release numbers for each subproject -- and then decide on a process to facilitate that need; I'm blanked out on its use at the moment. It seems to me we could completely replace individual project versions with the overall release ("17.6.0"), *if* we accept that non-changing projects would have several versions tagging exactly the same code state. The other way would be to bump the individual projects' versions upon each overall release, iff they have changes (at all? to their API?) since the last release. This would mean way more release work than just glorifying a given state of the master branch, but would make a lot of sense in the logic of a release process. The effort should hopefully be fine to do every 3 months, only. Your thoughts are welcome. Actual example implementations would also be welcome to provide a basis to work on. All in all we have agreed on various guidelines/facts but have not completely got a fix on the process and all the details yet. Please provide your feedback, and/but let's decide soon. ~N

Hi, I agree with Harald, adding the "net" part makes sense if we want to keep osmocombb out of it, "cellnet" sounds good. Not sure if adding a reference to "Infrastructure" could also make sense, such as "Osmo-CellInfra" or "Osmo-CellInfraNet". whenever >> a "new release (TM)" would be made ... but we never defined when we >> would do that nor how exactly. As it may be difficult or tedious to sync/fetch the changes from different project repositories, we could have a wiki page with a list of new added features for each release, and when new stuff is pushed into a given project they could add an entry in the changelist for that project for the next release. I have to say I'm not particularly against this idea from Neels. It would simplify a lot naming of the versions of all projects. -- - Pau Espin Pedrol &lt;pespin(a)sysmocom.de&gt; http://www.sysmocom.de/ ======================================================================= * sysmocom - systems for mobile communications GmbH * Alt-Moabit 93 * 10559 Berlin, Germany * Sitz / Registered office: Berlin, HRB 134158 B * Geschaeftsfuehrer / Managing Director: Harald Welte

Hi, Imho repo is the better choice compared to git submodule, which I so far only used in quite basic no-dependency projects to "bind"one common-ci repo. I made a spike/hands-on [1] last night to get a better understanding of repo, because I so far only consumed/used it. The created "repo" clones all necessary repositories for the accelerate3g5 CN stack, but the default.xml manifest already holds all Osmocom-CellNet repositories mentioned in this thread. So one can already switch between "default" and "accelerate3g5" project as well as add new projects. My understanding is that we need branches/tags to use repo for several Osmocom-CellNet projects (not optional). That's why I created "accelerate3g5" branch where release (candidate) tags of actual git repositories can be tested and tagged as release (candidate) within the "accelerate3g5" repo branch. The tested/verified state within the accelerate3g5 branch will be then merged to the master and tagged as a release to detach its state from HEAD. Inspired by your post I added a script to build the accelerate3g5 CN stack to the repo repository, so one can build the entire stack with 3 lines: $ repo init -u https://github.com/blobbsen/repo-handson -m accelerate3g5.xml $ repo sync $ .repo/manifests/build_accelerate3g5.sh Note: should work on FreeBSD too (gmake/make topic) Probably worth discussing is the topic whether mentioned build script should live in the repo repository or in osmo-ci (if we agree to add osmo-ci)? Although, osmo-ci sounds like a good place I like the fact one can put those build scripts also in the repo repository itself. Perhaps an osmo-repo helper script, which holds some handy shortcuts à la switchToProject, switchToBranch, because to switch branches or project one basically uses the three above mentioned lines again (cumbersome). Interesting idea! :) What do you think about [1] in general? Cheers, André [1] https://github.com/blobbsen/repo-handson

Hi Neels, On Wed, Apr 26, 2017 at 04:07:15PM +0200, Neels Hofmeyr wrote: actually, it's mostly the dashed version for the executables, and then the git repo follows that. I also like the shorter names. However, we already have the Osmocom trademark in US and EU, and it might make sense to reflect that in naming, should the need ever arise to defend against a copycat or some other misrepresentation of the project and/or its software. my vote goes for the latter, due to above reasoning. -- - Harald Welte &lt;laforge(a)gnumonks.org&gt; http://laforge.gnumonks.org/ ============================================================================ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6)