Otherwise, make reconfigures every time and then rebuilds all files.
I went too far in 3af9f5ed1a. All that's
needed is to make the configure target not use Build/SysConfig.hs as the
target name, so make won't delete that file after a failed build.
This commit was supported by the NSF-funded DataLad project
The problem with that target was, if a target like git-annex that
depended on it failed for some reason, make would delete
Build/SysConfig.hs, since it knows it's an intermediate file. But, since
stack only builds that file once, that caused all subsequent make git-annex
builds to fail.
Also, this avoids a double stack build when building with stack. Since
stack has no configure stage, and the Build/SysConfig.hs target was
about running the configure stage, the only way to only build once is to
combine the targets like this.
This should work better on the autobuilders that build with stack.
This commit was sponsored by NSF-funded DataLad project
Git does not provide a switch to find out where this directory is, and
while the git-init man page says it will always be in
/usr/share/git-core/templates, that's not the case on OSX with git
installed from homebrew. So, I used a hack taking the --man-path and
constructing a path from that. Works on both Debian and OSX at least.
Almost working, but there's a bug in the relaying.
Also, made tor hidden service setup pick a random port, to make it harder
to port scan.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
I've long considered the XMPP support in git-annex a wart.
It's nice to remove it.
(This also removes the NetMessager, which was only used for XMPP, and the
daemonstatus's desynced list (likewise).)
Existing XMPP remotes should be ignored by git-annex.
This commit was sponsored by Brock Spratlen on Patreon.
ghc 8.0.2 may make this unncessary, but it's not in a stackage version yet,
so put in a workaround.
Note that the linux builds already delete the RPATHs for similar reasons.
This commit was sponsored by Josh Taylor on Patreon.
The Makefile was putting them in git-annex.linux/i18n/i18n, and so I18NPATH
did not point to the files. I think that on close enough to Debian systems,
localedef then fell back to using the system-wide locale files, while on
other systems it would fail to generate locales.
Currently only done for utf-8 locales because the charset can easily be
told for those. Other locales don't include the charset in their name.
The locale definition is generated under git-annex.linux/locales.
So, this only works if the user can write there.
If locale generation fails for any reason, it's silently skipped.
The git-annex-standalone.deb installs the bundle under /usr, so this locale
generation won't work for non-root users.
This was originally done in a7ef05a9, but got lost in some change to the
Makefile. Use CROSS_COMPILE=Android to tell configure that it's configuring
for android instead of passing it a parameter.
This actually runs faster than building the man pages from the makefile
did. But the main purpose is to let Setup.hs import Build.Mans and so not
need the makefile.
The tarball on hackage will include only the files needed for cabal install;
it is NOT the full git-annex source tree. While it's totally obnoxious that
cabal files need every file listed out when basic wildcard support could
avoid hundreds of lines, and have to be maintained when files are added,
this does get the tarball size back down to 1 mb.
This also stops stack from complaining that it found modules not listed in
the cabal file.
debian/changelog, debian/NEWS, debian/copyright: Converted to symlinks
to CHANGELOG, NEWS, and COPYRIGHT, which used to symlink to these instead.
This avoids needing to include debian/ in the hackage tarball.
Setup.hs: Build man pages at install time using make and mdwn2man.
If it fails, which it probably will on windows, just skip installing
them.
I needed BUILDEROPTIONS to allow passing flags to stack build, but it also
lets me move the -j1 out of the normal build path, and to debian/rules
which has the goal of having a reproducible build
This was in the cabal file earlier, and was removed because it broke the
android cross build. Moving to the git-annex target of the Makefile
will make it be used for Debian packages etc but not android cross builds
or make fast or when users build with cabal.
As a result of the Makefile changes, the Debian package is built
with various hardening options. Although their benefit to a largely
haskell program is unknown.
This needs a patch to git to cause the git-annex completion to be
auto-loaded when completing "git annex <tab>". Otherwise, it will only
load when "git-annex" is tab completed. Once loaded, it works for both
uses. I've submitted the git patch to the git mailing list.
Also automated downloading the builds, finally. I had done it by hand until
now.
Note that the Windows autobuilder has an expired cert, so it will refuse to
download from it currently. I have emailed its admin to get that fixed,
hopefully.
This commit was sponsored by Peter Hogg.
Also re-add removed file, seem to work now.
hdiutil has *horrible* error reporting. You get back a number from 1 to
1000, with no indication of the problem. --verbose doesn't help.
So this is all guesswork.
The shell code was nasty, and buggy. New haskell code is much nicer,
and it's easy to do complicated calculations to properly convert possibly
absolute symlinks between libraries into relative links using it.
Was able to reuse many of the android patches, but several had to be
re-done. On Android, ghc is a stage2 build, so can compile, but not run TH
code. But debian's ghc on armel cannot even compile TH code, so it has
to be patched out.
Some haskell packages have been updated to new versions, including yesod
and DAV, and their patches had to be redone.
The Makefile now has 2 new targets. The first is run on a companion x86
system to do the build and get TH splices. Then the second target is run
the same source tree on the arm system to build without needing TH.
This commit was sponsored by Svenne Krap.
