After building an AUR package don't install it right away. Instead queue
it for install.
Then when a package comes along that does not have all of its
dependencies satisfied, install the queued packages before builting.
This allows a standard AUR update to be done with one transaction, while
building AUR packages with long dependency chains will work as normal.
This feature is behind the `--batchinstall` flag.
merge both dependency resolving parts into depSolver. The name was
choosen simply to not colide with the current code.
depSolver works but it not implemented in the install process and is
missing conflict checking.
Yay first calls 'makepkg --nobuild -fC' to update the pkver. Later on we
call 'makepkg -cf --noectract --noprepare --holdver' to actually build
the package.
Inbetween these two calls we keep the already extracted
sources to save time on the reextract and duplicated call to prepare
(pkgbuilds should not require user input but things such as linux-ck do
and calling prepare twice will actually cause them to promt twice)
We also have two checks. First we see if the package is already
installed and up to date (--needed) and secondly we check if the
package is already built.
If any of these conditions are met we skip building the package. This
leaves a dangling src/ directory as 'makepkg -c' was never ran.
Now if these conditions are met tell makepkg to cleanup and exit.
Before --needed was left purerly to pacman. The problem with this is
that if a package is not in the cache, it will end up being build just
for pacman to skip over the install.
Now Yay will handle this and skip the build and install if --needed is
used. The inital clone and pkgver bumb is still donw.
There was a possibly that getIncompatible could be broken when working
with split packages. After looking into it, this does not seem to be the
case, as long as we only work with pkgbases.
arch may be overriden inside of a package function. The global arch
array may be 'any' but package_foo could say arch=('x86_64'). If you
were running arm then should yay warn about this?
I have decided the answer is no. This is because makepkg doesn't care.
It would seem makepkg only checks the global arch array. This leads me
to believe that the global array should contain a the required arch.
Therefore the correct way to do the former example would be to declare
the global arch as 'x86_64' and then override package_bar with 'any'.
So all that we should do is properly use formatPkgbase to list all the
packages under the base.
depOrder.Aur contains the order in which AUR packages are to be
installed. While depOrder.bases contains the actual package data
organized by pkgbase.
deoOrder.AUR is kind of counterintuitive, as it only contains one
package from each package base.
For example if you were to install libc++{,abi,experimental},
depOrder.Aur would only contain one of those packages, which one
actually being quite random. depOrder.Bases[pkg.Pkgbase] will then be
looked up and everything under that slice would be installed.
This means that the only real use depOrder.Aur has, is to give the
pkgbase. So to cut out the middleman, lets merge .Aur and .Bases into
a single field.
Doing this has also heped to spot som subtle bugs:
Fix subtle split package errors.
The number menus now correctly list and respect (installed) for bases
that have atleast one of their packages installed.
Entering package names in number menus now always expects the pkgbase
name instead of the random package which happened to make it into .Aur.
--rebuild and --redownload correctly handles split packages.
formatPkgbase is also used more.
Only pass the packages belonging to the desired pkgbase. Before our
entire list of every pkgbase had to be passed, as well as an example
package from the base we are looking for.
This lets use use formatPkgBase easier in other places outside of
install.
Many of the functions in install could also be simplified in a similar
way, although that has not been done in this commit.
Yay's dependency resolving takes provides into account. When upgrading
AUR package 'foo', if a repo package provides 'foo' then yay would get
confused and pull in the package providing 'foo' instead of the AUR
package.
This commit ensures AUR upgrades always exclusively check the AUR.
Bash seperates on whitespace, so the fish completion file
actually works for bash and zsh. So remove the concept of shells
entirley and just use the singular aur_sh.cache file.
If for some reason output without the repository data is needed, the
user could always just pipe it into awk like so
`yay -Pc | awk '{print $1}'`. Or perhaps a --quiet option could be added
where yay will strip the output itself.
The completion cache now updates when installing AUR packages. This is
done as a goroutine with no wait groups. This ensures the program will
never hang if there is a problem.
The completion is stil updated during -Pc but as long as an AUR package
has been installed recently it should not need to update.
The cache will now also wait 7 days instead of 2 before refreshing.
A refresh can be forced using -Pcc.
When using nocombinedupgrade "there is nothing to do" is printed when
insinstalling repo packages. This is because as far as AUR
installer is concerned, there is nothing to do. Instead only print that
when doing a sysupgrade.
The main reason behind this is for future localisation. yes and no can
be set to the localized equivalent and it should just work.
This Refactor also changes the function in ways which make it much less
confusing.
The param is no longer reversed and changed to a boolean. Before you had
to pass in Yy to get a default of no and vice versa.
The function now retuens false for no and true for yes. Before it would
return true if the default option was choosen.
Before:
* Empty line after `Searching AUR`
* `There is` starts with capital letter without space before
```
> yay --combinedupgrade -Syu
:: Synchronizing package databases...
core is up to date
extra is up to date
community is up to date
multilib is up to date
:: Searching databases for updates...
:: Searching AUR for updates...
:: Starting full system upgrade...
there is nothing to do
> yay --nocombinedupgrade -Syu
:: Synchronizing package databases...
core is up to date
extra is up to date
community is up to date
multilib is up to date
:: Starting full system upgrade...
there is nothing to do
:: Searching databases for updates...
:: Searching AUR for updates...
There is nothing to do
```
After:
```
> yay --combinedupgrade -Syu
:: Synchronizing package databases...
core is up to date
extra is up to date
community is up to date
multilib is up to date
:: Searching databases for updates...
:: Searching AUR for updates...
:: Starting full system upgrade...
there is nothing to do
> yay --nocombinedupgrade -Syu
:: Synchronizing package databases...
core is up to date
extra is up to date
community is up to date
multilib is up to date
:: Starting full system upgrade...
there is nothing to do
:: Searching databases for updates...
:: Searching AUR for updates...
there is nothing to do
```
When looking at diffs of packages downloaded as tar achives actually
show a diff instead of opening the files in the ediror. This diff
is against /var/empty so it is not that useful. In realiy this is an
excuse to move the srcinfo parsing back down to after the git merge.
Viewing the build files in the editor requires the .srcinfos to be
parsed to know what .install files are used. Now viewing diffs does not
need the srcinfos so they can be moved to after we git merge.
Before now the srcinfo would have been of the previous version. This is
not much of a problem because we don't really use the srcinfo for much.
Checking the arch and pgpkeys which never really change.
Recently libc++ changed their pgp keys and yay missed that because it
parsed the old srcinfo without the new keys.
Viewing a proper diff for tars can be tossed on the todo by doing
something along the lines of:
mv pkg{,.old}
diff pkg{,.old}
rm -rf pkg.old
But I doubt there are many people out there using tar so it's not much
of an issue.
Previously each call to an external command had two functions.
PassToFoo() and PassToFooCapture(). These functions are always similar
and end up with duplicated code.
So instead have the passToFoo() functions return the cmd itself and
create small helper functions show() and capture() which will run the
command and either forward it to std{out,err,in} or capture the output
Also the saveVCSInfo() function which was called after every makepkg
call is now only called after the pacman -U succeeds.
Currently When performing a system upgrade, Yay will first refresh the
database then perform the repo and AUR upgrade. This allows Yay to add
some features such as better batch interaction, showing potential
dependency problems before the upgrade starts and combined menus
showing AUR and repo upgrades together.
There has been discussion that this approach is a bad idea. The main issue
people have is that the separation of the database refresh and the upgrade
could lead to a partial upgrade if Yay fails between the two stages.
Personally I do not like this argument, there are valid reasons to Yay
to fail between these points. For example there may be dependency or
conflict issues during the AUR upgrade. Yay can detect these before any
installing actually starts and exit, just like how pacman will when
there are dependency problems.
If Yay does fail between these points, for the previously mentioned
reasons or even a crash then a simple refresh will not cause a
partial upgrade by itself. It is then the user's responsibility
to either resolve these issues or instead perform an upgrade using
pacman directly.
My opinions aside, The discussions on the Arch wiki has reached
a decision, this method is not recommended. So to follow the decided
best practises this behaviour has been disabled by default.
This behaviour can be toggled using the --[no]combinedupgrade flag
It should be noted that Yay's upgrade menu will not show repo packages
unless --combinedupgrade is used.
--ask is no longer used when installing AUR packages, instead pass no
confirm when we know there are no conflicts and wait for manual
confirmation when there are.
This means that when there are no conflicts there should be no change in
behaviour and the user will not need to intervene at all.
The old behaviour can still be used with --useask.
Instead of doing all the AUR stuff just pass to pacman and return. No
need for any of Yay's stuff when there's no AUR involved.
Of couse everything before that still happens. Upgrade menu ect.
Clean build needs to happen before downloading pkgbuilds so that they
can be deletd before downloading.
Editing and diff viewing needs to happen after downloading the
pkgbuilds.
Prevously we asked to clean and edit at the same time. Then clean,
download pkgbuilds and open the editor.
This poeses a problem for diff viewing and editing. It's likley that the
user will see the diff and use that to decide if they want to edit the
pkgbuild. Using the current method, the user will be asked to view diffs
and edit before actually seeing any diffs.
Instead split cleaning diff showing and editing to three seperate menus
in the following order:
show clean menu
clean
download pkgbuilds
show diff menu
show diffs
show edit menu
edit pkgbuilds
Also each menu is seperatly enableable. By default only the diff menu is
shows. If the user wishes to clean build, edit pkgbuilds or disable
diffs then the user can use the --[no]{clean,diff,edit}menu flags. This
replaces the --[no]showdiffs flags.
Newly cloned packages already start out at origin/master, so there is no
diff to show. Track if we cloned a package and if so make sure to show
the full diff
This is what 5775e3..43d2a6 has been leading up to. Git fetch will be
called on all pkgbuilds, then the user is offered a chance to view the
diffs. If they choose to continue, merging happens. This allows users to
abort the install after viewing diffs and still be able to see thoes
diffs again if they try to install later on.
This also makes the git stuff a little more modular which should help in
organzing diff showing + pkgbuild editing.
The order of targets does somewhat matter. For example doing something
like 'pacman -S db1/foo db2/foo' should cause the second package to be
skipped.
The order of targets also effects in which order they are resolved. This
should make errors more reproducable if any ever occur.
Prepare ends up getting ran twice every time we install a package,
theres not problems with doing so apart from a little inefficiency.
Previously the install flow would be like this:
downlod sources + verify
prepare + pkgver bump
full build (prepare included)
Now on the last point pass no extract to use the srcdir from the
previous command and pass noprepare and holdver because we allready did
these steps previously.
diff viewing can be toggled via --[no]showdiffs. When enabled diffs will
be shown for packages between the current HEAD and upstream's HEAD.
Packages downloaded via tarballs will be shown in full using the editor
git diff is used to show diffs. Therefore the pager for diffs can be
set via the PAGER and GIT_PAGER enviroment variables.
These flags limit operations to only check the repos or only check the
AUR. These flags apply to -S, -Si and -Su.
-a may also be used as a short option for --aur. --repo has no short
option as -r is taken.
Pacman 5.1 removes the symlink to the current directory for built
packages. This causes Yay to break for people who have set an external
PKGDEST.
Pacman 5.1 also brings an improved --packagelist option. This makes
it much simpler to find where packages will be placed. Hence this fix
also simplifies the code.
Yay has an -Sc option to clear it's cache. If using an external PKGDEST
this is now mostly useful for clearing out old pkgbuilds and sources.
paccache should be used for cleaning build packages.
The previous warning system would show warnings recursivley for all
packages being resolved. While I like this, other have complained at it
being overly verbose.
Either way the main purpose of this is to allow warnings to be printed
before the upgrade menu shows. This is mostly just to get a usable
warning system.
This may change if a better solution is found.
Targets are used for tracking wether a package should be marked as
explicitly installed or as a dependency. This is not ideal because you
can have a target such as java-environment that resolves to a different
package.
Therefore Targets are now used only for the initial dependency resolving
and checking for missing dependencies. The Explicit set is now used to
mark what packages are explicit, seperate from the targets.
The warnings were moved down to after the upgrade menu, mainly because
it is a lot easier to do this way, it may get moved back if it can be
done in a non hacky way,
Sort the provider menu alphabetically. Always ensure direct matches show
up first. This ensures hitting enter for the default value will always
be the same package that the user/dependency requested if an exact match
exists.
If a package is already installed pick that instead of providing a menu.
Ensure duplicates do not show up in the menu.
MakeOnly would be set to true when moving from normal deps to make deps
But would incorrectly stay set to true when moving to the deps of the
following packages.
depOrder.Aur now only holds one package from each base like
depCatagories does.
If --ignore was specified on the command line and the user skips
packages using the number menu, packages would not be properly skipped
because they the manual --ignore would overide the --ignore from the
menu.
Now correctly combine both --ignore flags into a single combined flag
when passing to pacman.
