Releases: mvx-lang/mv_git
Releases · mvx-lang/mv_git
Release list
git 2.1.0-rc3
What changed since 2.1.0-rc2
All ports
- only a file's records are the sweep's to prune (#249)
- an ordinary tracked directory is not a deleted MV file (#247)
- build-jbase.sh: link the antlr4 runtime jBASE's own libjQL.so needs (#245) (#244)
- keep any backend's store out of a commit, not just lmdb's (#240)
UniData
- udt: never ship a CallC fragment without its objects (#238)
jBASE
git dev
What changed since 2.1.0-rc2
All ports
- only a file's records are the sweep's to prune (#249)
- an ordinary tracked directory is not a deleted MV file (#247)
- build-jbase.sh: link the antlr4 runtime jBASE's own libjQL.so needs (#245) (#244)
- keep any backend's store out of a commit, not just lmdb's (#240)
UniData
- udt: never ship a CallC fragment without its objects (#238)
jBASE
git 2.0.4
git 2.1.0-rc2
git 2.1.0-rc1
Same tree as 2.1.0-beta6, promoted. Nothing has changed since; the six betas are what this asks about.
2.1 is the jBASE port and the two ports that could not install cleanly learning to. Twenty-nine issues since 2.0.3.
Compatibility, unchanged from 2.0: the control format changed in 2.0, and a 2.x repository must not be opened with 1.6.x.
The jBASE port — #114
jBASE joins MVX, UniData and UniVerse as a shipping port, with its own binary artifact.
- #125
jb-gitgainsadopt - #177 thirteen programs used jBASE reserved words as identifiers
- #179
GIT.OBJhad no jBASE branch for most git-object operations - #180 the suite read
MVXGITon the verb arm, where it is not set - #183
GIT ATTRread file geometry that does not exist there - #184
GIT CLONEspent a second licence seat handing over tojb-git - #175 LINK swallowed the catalog output, so failed compiles were invisible
- #201
mvpkg.jsondid not declarejbase, so mvpkg would not install mv_git there - #203 dictionary associations were not remapped
- #204 default column lists (
_PH_/&PH&phrases) did not cross platforms
Installable by the package manager
Two ports could not be installed cleanly by mvpkg. Both can now — verified end to end from published artifacts, with a released client.
- #212 UniData: an mvpkg install could not compile at all. Every program opens
$INCLUDE BP.INC PLATFORM.H, and only the package's owninstall.shregistered that name — which an mvpkg install never runs. The package declares the include now (needs mvpkg ≥ 1.22.0). - #224
GIT.RPChad never compiled on UniData. It is UniVerse's transport, guarded "not MVX" — which is not the set that hasWRITEBLK/READBLK. The installer also sent the compiler's output to/dev/null, so a failed compile left the old object in place and reported success; it now keeps the output, stops on a failure, and no longer feeds UniData's own compiled objects back to the compiler. - #226 / #208 UniVerse could not repair an account, only build one. A file there is a VOC record and the directories behind it, and those come apart. The install answered "already a file, nothing to do" for a name that named nothing; where it had deleted the directories itself,
CREATE.FILErefused to re-make them, so the state was unrecoverable by the code that made it. It repairs now, and reports whatCREATE.FILEsaid when it cannot. - #187, #165 the UniVerse suite's races with the two-seat licence, and the three different wrong ways it reported them
Asking mv_git what it is
- #215
GIT VERSION/GIT --VERSIONcould not answer outside a repository — the first thing anyone types after installing — and the UniData in-session engine reported no version at all - #218 a branch name is not a version:
$GITHUB_REF_NAMEis set on every Actions run, so builds off a tag stamped themselvesmain, in the binaries and the shipped manifests - #221 2.1.0-beta4's UniData binary shipped as
mv_git 0; a release that cannot name itself now fails instead
Correctness and the repository's own furniture
- #161 the repository still carried furniture it now refuses to stage
- #163 UniData:
addandstatusdisagreed about a record's trailing empty attribute - #168 the verb record is derived; declared in
VERBSinstead - #171 the master dictionary's
F*records travel, and they are the platform's - #172 the fresh-account assertion counted records that legitimately travel
- #189 libgit2's ownership check refused the operator's account
- #192 a clone that could not run was reported as success
- #164 udt-callc: a duplicate object is fatal now, not a warning
Release engineering
- #159 CI on every merge to main/develop, and no release unless every port is green
- #209 concurrent jobs raced to create the release; the loser left a draft
Artifacts: MVX, UniData, UniVerse, jBASE, and source.
git 2.1.0-beta6
What changed since 2.1.0-beta5
All ports
- Merge pull request #231 from mvx-lang/develop (#231)
- Merge pull request #229 from mvx-lang/develop (#229)
- Merge pull request #227 from mvx-lang/develop (#227)
UniData
- the installer fed UniData's own objects back to the compiler (#224)
UniVerse
- the install has to repair an account, not only build one (#226)
git 2.1.0-beta5
git 2.1.0-beta4
git 2.1.0-beta3
git 2.1.0-beta2
What changed since 2.1.0-beta1
All ports
- Merge pull request #207 from mvx-lang/develop (#207)
- %PH% is the canonical default column list, @ is the platform's (#204)
- Make mv_git installable by mvpkg: declare jbase, the artifact base, and a valid jBASE key (#201)
UniVerse
- assert the preconditions of the fast-forward pull (#187)