[pacman-dev] [PATCH] makepkg --packagelist: just list the built package files we will build
Currently this seems to be only theoretically useful. The most likely
reason for wanting a packagelist is in order to script makepkg and
derive the filenames for the packages we want to install or repo-add,
but in the current implementation this requires a lot of additional
post-processing which must be duplicated in every utility to wrap
makepkg.
- It is of minimal use to know what packages might get created on some
other device utilizing a different CPU/OS architecture, so don't list
them.
- It is non-trivial to reimplement makepkg's logic for sourcing any of
several makepkg.conf configuration files, then applying environment
overrides in order to get the PKGDEST and PKGEXT, so include them
directly in the returned filenames.
Signed-off-by: Eli Schwartz
On 31/10/17 04:37, Eli Schwartz wrote:
Currently this seems to be only theoretically useful. The most likely reason for wanting a packagelist is in order to script makepkg and derive the filenames for the packages we want to install or repo-add, but in the current implementation this requires a lot of additional post-processing which must be duplicated in every utility to wrap makepkg.
- It is of minimal use to know what packages might get created on some other device utilizing a different CPU/OS architecture, so don't list them. - It is non-trivial to reimplement makepkg's logic for sourcing any of several makepkg.conf configuration files, then applying environment overrides in order to get the PKGDEST and PKGEXT, so include them directly in the returned filenames.
Signed-off-by: Eli Schwartz
Needs makepkg man page update too.
--- scripts/libmakepkg/util/pkgbuild.sh.in | 12 +++++------- 1 file changed, 5 insertions(+), 7 deletions(-)
diff --git a/scripts/libmakepkg/util/pkgbuild.sh.in b/scripts/libmakepkg/util/pkgbuild.sh.in index 2a4bd3af..6cfda0b3 100644 --- a/scripts/libmakepkg/util/pkgbuild.sh.in +++ b/scripts/libmakepkg/util/pkgbuild.sh.in @@ -149,14 +149,12 @@ print_all_package_names() { local version=$(get_full_version) local architecture pkg opts a for pkg in ${pkgname[@]}; do - get_pkgbuild_attribute "$pkg" 'arch' 1 architecture + architecture=$(get_pkg_arch $pkg) get_pkgbuild_attribute "$pkg" 'options' 1 opts - for a in ${architecture[@]}; do - printf "%s-%s-%s\n" "$pkg" "$version" "$a" - if in_opt_array "debug" ${opts[@]} && in_opt_array "strip" ${opts[@]}; then - printf "%s-%s-%s-%s\n" "$pkg" "@DEBUGSUFFIX@" "$version" "$a" - fi - done + printf "%s/%s-%s-%s%s\n" "$PKGDEST" "$pkg" "$version" "$architecture" "$PKGEXT" + if in_opt_array "debug" ${opts[@]} && in_opt_array "strip" ${opts[@]}; then + printf "%s/%s-%s-%s-%s%s\n" "$PKGDEST" "$pkg" "@DEBUGSUFFIX@" "$version" "$architecture" "$PKGEXT" + fi done }
Currently this seems to be only theoretically useful. The most likely
reason for wanting a packagelist is in order to script makepkg and
derive the filenames for the packages we want to install or repo-add,
but in the current implementation this requires a lot of additional
post-processing which must be duplicated in every utility to wrap
makepkg.
- It is of minimal use to know what packages might get created on some
other device utilizing a different CPU/OS architecture, so don't list
them.
- It is non-trivial to reimplement makepkg's logic for sourcing any of
several makepkg.conf configuration files, then applying environment
overrides in order to get the PKGDEST and PKGEXT, so include them
directly in the returned filenames.
- Output is null-delimited, in order that the filenames can be reliably
parsed.
Signed-off-by: Eli Schwartz
On 06/02/18 12:47, Eli Schwartz wrote:
Currently this seems to be only theoretically useful. The most likely reason for wanting a packagelist is in order to script makepkg and derive the filenames for the packages we want to install or repo-add, but in the current implementation this requires a lot of additional post-processing which must be duplicated in every utility to wrap makepkg.
- It is of minimal use to know what packages might get created on some other device utilizing a different CPU/OS architecture, so don't list them. - It is non-trivial to reimplement makepkg's logic for sourcing any of several makepkg.conf configuration files, then applying environment overrides in order to get the PKGDEST and PKGEXT, so include them directly in the returned filenames. - Output is null-delimited, in order that the filenames can be reliably parsed.
Signed-off-by: Eli Schwartz
--- v2: use null delimited text, update manpage with the effects of the patch.
Newline separation should be fine. makepkg ensures most (all?) of the fields are not allowed to have newlines in them. If any can, we need to catch that in out PKGBUILD linting and error out. Probably need to add a check that the fields from makepkg.conf don't have newlines either. A
On 02/07/2018 07:24 PM, Allan McRae wrote:
Newline separation should be fine. makepkg ensures most (all?) of the fields are not allowed to have newlines in them. If any can, we need to catch that in out PKGBUILD linting and error out. Probably need to add a check that the fields from makepkg.conf don't have newlines either.
I was thinking primarily of the fields in makepkg.conf, since AFAICT no PKGBUILD fields used in assembling the filename can contain newlines. OTOH, both PKGDEST and PKGEXT currently allow this, as do all the other directory variables in makepkg.conf Should we add a lint_pkgbuild check for all the makepkg.conf variables? -- Eli Schwartz Bug Wrangler and Trusted User
On 08/02/18 11:22, Eli Schwartz wrote:
On 02/07/2018 07:24 PM, Allan McRae wrote:
Newline separation should be fine. makepkg ensures most (all?) of the fields are not allowed to have newlines in them. If any can, we need to catch that in out PKGBUILD linting and error out. Probably need to add a check that the fields from makepkg.conf don't have newlines either.
I was thinking primarily of the fields in makepkg.conf, since AFAICT no PKGBUILD fields used in assembling the filename can contain newlines.
OTOH, both PKGDEST and PKGEXT currently allow this, as do all the other directory variables in makepkg.conf
Should we add a lint_pkgbuild check for all the makepkg.conf variables?
Yes please. Although, lint_config.sh might be a better place to put it, as it is not really part of pkgbuild linting. Allan
On 02/07/2018 10:37 PM, Allan McRae wrote:
On 08/02/18 11:22, Eli Schwartz wrote:
On 02/07/2018 07:24 PM, Allan McRae wrote:
Newline separation should be fine. makepkg ensures most (all?) of the fields are not allowed to have newlines in them. If any can, we need to catch that in out PKGBUILD linting and error out. Probably need to add a check that the fields from makepkg.conf don't have newlines either.
I was thinking primarily of the fields in makepkg.conf, since AFAICT no PKGBUILD fields used in assembling the filename can contain newlines.
OTOH, both PKGDEST and PKGEXT currently allow this, as do all the other directory variables in makepkg.conf
Should we add a lint_pkgbuild check for all the makepkg.conf variables?
Yes please. Although, lint_config.sh might be a better place to put it, as it is not really part of pkgbuild linting.
Sure, anything else that we should lint while we are at it? I think mostly the only things to check are variable types, which we could probably crib from lint_pkgbuild/variable.sh -- Eli Schwartz Bug Wrangler and Trusted User
On 08/02/18 13:44, Eli Schwartz wrote:
On 02/07/2018 10:37 PM, Allan McRae wrote:
On 08/02/18 11:22, Eli Schwartz wrote:
On 02/07/2018 07:24 PM, Allan McRae wrote:
Newline separation should be fine. makepkg ensures most (all?) of the fields are not allowed to have newlines in them. If any can, we need to catch that in out PKGBUILD linting and error out. Probably need to add a check that the fields from makepkg.conf don't have newlines either.
I was thinking primarily of the fields in makepkg.conf, since AFAICT no PKGBUILD fields used in assembling the filename can contain newlines.
OTOH, both PKGDEST and PKGEXT currently allow this, as do all the other directory variables in makepkg.conf
Should we add a lint_pkgbuild check for all the makepkg.conf variables?
Yes please. Although, lint_config.sh might be a better place to put it, as it is not really part of pkgbuild linting.
Sure, anything else that we should lint while we are at it? I think mostly the only things to check are variable types, which we could probably crib from lint_pkgbuild/variable.sh
We should only lint things that we want to ensure match assumptions. For this patch, checking CARCH and PKGEXT are strings and don't contain newlines is enough. A
Currently this seems to be only theoretically useful. The most likely
reason for wanting a packagelist is in order to script makepkg and
derive the filenames for the packages we want to install or repo-add,
but in the current implementation this requires a lot of additional
post-processing which must be duplicated in every utility to wrap
makepkg.
- It is of minimal use to know what packages might get created on some
other device utilizing a different CPU/OS architecture, so don't list
them.
- It is non-trivial to reimplement makepkg's logic for sourcing any of
several makepkg.conf configuration files, then applying environment
overrides in order to get the PKGDEST and PKGEXT, so include them
directly in the returned filenames.
- Output is delimited by newlines, for readability. For maximum
parsing reliability, libmakepkg needs to learn how to lint the PKGDEST
and PKGEXT variables to ensure they do not contain newlines, which
will be submitted in a separate patch.
Signed-off-by: Eli Schwartz
participants (2)
-
Allan McRae
-
Eli Schwartz