Changes since 20.1.5:
Category Makefiles must only list subdirectories in SUBDIR that actually
contain a package. There is no need to mention the other directories.
This is the same as in the top-level Makefile, where mk/ and regress/ are
not listed.
Packages from pkgsrc-wip may have a COMMIT_MSG file with a suggested
commit message for importing the package into main pkgsrc.
This allows to show the actual changes by setting SUBST_SHOW_DIFF=yes,
and with SUBST_NOOP_OK=no it also demonstrates that no substitutions are
needed for substplistbasedirs.
Another thing this demonstrates is this wrong patch:
-echo >>$sedrules "s|@@PKGVERSION@@|@PKGVERSION@|g"
+echo >>$sedrules "s|@3.17@|3.17|g"
Changes since 20.1.4:
No more wrong warnings about the Solaris /bin/sh. These warnings had been
there for 14 years, preventing pkgsrc developers from using the $$(...)
command substitution and negation in shell conditions.
https://mail-index.netbsd.org/pkgsrc-changes/2020/05/01/msg212194.html
Changes since 20.1.3:
For patches that patch a single file, the filename of the patch should
correspond to the patched file. There are a few different naming schemes
in action, therefore the check is relatively loose. Patches that are
called patch-[a-z][a-z] continue to be allowed for historic reasons.
Patches that are called patch-CVE-* are also allowed.
The entries in doc/CHANGES-* are checked for consistency. For example,
it doesn't make sense to add a package twice or "update" a package from
version 1.0 to version 1.0. All version numbers in these entries must
be valid pkgsrc versions, i.e. start with a digit and only use
characters from -.0-9A-Z_a-z.
Changes since 20.1.2:
Stricter check for Python version numbers. Before, 25 and 26 had not
been marked as wrong.
In assignments like PKGNAME=${DISTNAME:S,from,to,}, modifiers that don't
have any effect generate a note. Most of these modifiers are redundant
or outdated.
Patches must not add well-known absolute paths like /usr/pkg, /var and
/etc since these must be overridable by the pkgsrc user. Other absolute
paths continue to be allowed.
The "jperkin should have run the test suite before releasing 0.16.0" release.
## Version 0.16.1 (2020-04-29)
* Don't attempt to update the keep list if we're only downloading packages.
* Avoid C99isms for now, fixes CentOS 6 build.
* Ignore "pkg_admin rebuild-tree" output, removes spurious "Done." messages
from output.
* Some minor internal refactoring.
PKGPATH is one of the few variables that is guaranteed to be defined when
mk.conf is included. Conditions like ".if ${PKGPATH} == category/package"
trigger a "Malformed conditional" when this variable is not defined.
## Version 0.16.0 (2020-04-27)
* Rebuild +REQUIRED_BY files after each pkgdb modification, ensuring
consistency and avoiding "Can't open +CONTENTS of ..." errors during
subsequent upgrades.
* Improved output sorting.
* Turn off WARNS.
* Minor cleanups.
* Documentation fixes.
* Flush yes/no prompt before reading response (Michael Forney).
Changes since 20.1.1:
Ensure that relative paths to other packages have the canonical form
../../category/package.
Add notes about pathname patters that mention ${WRKSRC} even though they
are already defined to be relative to WRKSRC.
Fix check for redundantly added categories. The check had previously
reported that the included file would be redundant, which was wrong.
It's always the including file that provides the redundancy.
The changes since version 1.99.4 appear to be:
- Fix pkg register issues with some version of libarchive:
* hardlinks not being made hardlinks sometime
* issues when PREFIX is a symlink to another directory
- Use portable mechanism to find the number of CPUs which allows pkg repo to be
faster on linux
Changes since 20.1.0:
In UNLIMIT_RESOURCES, the recently added virtualsize is allowed.
Packages that have distfiles without any digit in their name should
define DIST_SUBDIR to avoid polluting the global namespace. The
top-level distfiles directory should only contain versioned filenames.
Changes since 19.4.13:
It is an error to have TODO lines in DESCR files. These are typically
generated by url2pkg and should be replaced with proper text before
committing the package.
Changes since 19.4.12:
Files that are mentioned redundantly in PLIST files generate an error.
Missing DESCR files generate an error.
Hard-coded /usr/pkg in patches generates an error.
Changes since 19.4.11:
Redundant additions to BUILDLINK_API_DEPENDS and BUILDLINK_ABI_DEPENDS
get warnings since they may have been needed in the past but the
dependent package has increased its required version numbers over time.
Changes since 19.4.10:
The use of PKG_OPTIONS and PKG_BUILD_OPTIONS in buildlink3.mk and other
files is checked for common mistakes.
Checking the indentation of a continuation line no longer crashes in edge
cases.
Changes since 19.4.0:
* Makefile.am and Makefile.in are checked even though they don't start
with a #! line.
* Only shell programs with an interpreter that is clearly a POSIX shell
are checked. Before, there was a blacklist of interpreters to be
skipped.
* Lots of refactorings and code cleanups.
* Some additional test files.
Checks whether the given files use features of programming languages that
are not portable to a wide range of platforms.
See mk/check/check-portability.mk.
https://mail-index.netbsd.org/tech-pkg/2020/03/10/msg022787.html
This new check is not active by default, and the package is not forced to
be installed. This will be done after getting some experience in
practical cases.
The code has been tested by extracting about 1.7 GB of pkgsrc distfiles
and scanning for lines containing both "[[" and "]]".
The function make.df.depends() constructs a dataframe for DEPENDS lines in
the updated Makefile. That dataframe is combined with other dataframes for
different portions of the Makefile. Consequently, all the dataframes must
have the same fields or combining with rbind() fails. Previously, the
dataframe had the wrong set of fields; this explicitly selects the correct
set that matches the other dataframes so that they can all be combined
correctly.
Changes since 19.4.9:
In continuation lines with long values, pkglint no longer suggests to
move the continuation backslash in the middle of the variable value, as
that would be impossible.
Warn when a shell command is assigned to a variable that only takes
pathnames. Shell commands can contain command line options, and these
are not pathnames.
The TOOLS_PLATFORM.tool variables are not defined on every platform.
When these variables are used outside an OPSYS check, a warning lists
the platforms where the tool is undefined or only defined conditionally.
In order for recursive calls of R2pkg to work, they must receive the same
set of arguments as the original. Previous argument handling stripped the
hyphen (-) from each argument; this explicitly restores it.
Changes since 19.4.8:
Packages that include omf-scrollkeeper.mk even though their PLIST doesn't
contain any .omf files will generate an error message, suggesting that
the .include line be removed.
Changes since 0.5:
mktemp does not use a file pattern anymore. That file pattern had the
effect that all temporary files had the same name: /tmp/pkg_regress.
Changes since 19.4.7:
The diagnostic for homepages using FTP is simpler now.
When running pkglint recursively on the top-level directory, the
inter-package checks (distfile hashes, unused licenses) are enabled
implicitly. This way, the only effect of the -Cglobal option is now
whether the pkgsrc infrastructure files are checked as well.
The check for removed packages that have not been recorded in
doc/CHANGES prints the correct lines when pkglint is run with the
--source option.
Fatal technical errors are no longer treated as diagnostics since they
are none. That was an early conceptual mistake, but since these fatal
error didn't happen often, it didn't matter.
In diagnostics, when referring to other lines, the previously used words
before/after have been replaced with above/below to avoid any confusion
whether space or time is meant.
In CONF_FILES, spaces and quotes are allowed.
See https://gnats.netbsd.org/42191.
Fixed unintended side-effects when running pkglint --autofix --only.
Before, all fixes were applied to the file, whether or not they matched
the --only option.
Fixed resolution of relative paths of the form ../../category/package
when they appeared in an infrastructure file.
Lots of refactorings and housekeeping, as usual.
Changes since 19.4.6:
HOMEPAGE definitions that use http no longer get a warning that they
should migrate to https. Those that could be migrated have been migrated,
and the remaining homepages are not yet ready, so there's no benefit in
having this warning by default. Only in --network mode and when the https
site is indeed reachable, the warning is shown.
In long continued lines, the continuation backslash should be preceded by
a single space. Pkglint had wrongly removed that space in a few cases
before.
pkglint -r --network --only "migrate"
As a side-effect of migrating the homepages, pkglint also fixed a few
indentations in unrelated lines. These and the new homepages have been
checked manually.
Changes since 19.4.5:
Added the --network option, which allows pkglint to use HTTP calls for
determining whether the homepage of a package is reachable.
Added migration from http URLs to the corresponding https URLs. This is
only done if the https URL is indeed reachable or well-known to support
https.
Added migration from https SourceForge URLs back to http URLs since a
previous pkglint run had migrated URLs to non-working https URLs. See
https://mail-index.netbsd.org/pkgsrc-changes/2020/01/18/msg205146.html.
Added a warning for HOMEPAGE that uses ftp:// since that is not
user-friendly. In the same way, download-only host names on SourceForge
are not suitable as a homepage since they usually only generate a 404.
Changes since 19.4.4:
Fixed automatic replacement from http SourceForge URLs to their https
counterparts. According to their official documentation, the https URLs
are not on the same hosts as before, but on *.sourceforge.io.
For those of us who have multiple pkgsrc trees, or move them around,
and forget the path gets hard-coded into the shell script, or make a
typo when supplying an alternate PKGSRCDIR, and then wonder why nothing
gets cleaned.
pkglint --only "https instead of http" -r -F
With manual adjustments afterwards since pkglint 19.4.4 fixed a few
indentations in unrelated lines.
This mainly affects projects hosted at SourceForce, as well as
freedesktop.org, CTAN and GNU.
Changes since 19.4.2:
PLIST_VARS identifiers must not contain characters that are interpreted
specially in regular expressions.
All pkgsrc text files except for doc/pkgsrc.{html,txt} must use the
default CVS keyword substitution.
Changes since 19.4.1:
Fixed variable resolution. Before, variables that had not been defined
in an included file could still end up in the scope of the package,
which made many of the pkglint checks unreliable.
Each condition that is used in the PLIST should be defined somewhere in
the package Makefile or its included files.
When loading a package Makefile, hacks.mk is loaded implicitly at the
end, just as the pkgsrc infrastructure does in mk/bsd.hacks.mk.
Changes since 19.4.0:
The notes for inserting an empty line have been changed from "insert
after this line" to "insert before this line" to make the line numbers
in the diagnostics contiguous. There had been several places where the
diagnostics went from line 1 to line 2 and then back to line 1, which
was confusing.
The lines in ALTERNATIVES files are checked for trailing whitespace.
This is only for consistency with the other checks. In the whole pkgsrc
tree all ALTERNATIVES files are already fine.
The diagnostics for comments in .endif/.endfor lines that don't
correspond to their .if/.elif/.for counterparts now includes the exact
line number of the corresponding condition, to make the warning easier
to fix.
The diagnostics for wrong variable value alignment now mention the
current column in addition to the desired column, to make it easier to
see by how much and in which direction the indentation should be fixed.
Variables that are used in conditions before they are actually defined
need the :U modifier.
Changes since 19.3.19:
Empty PLIST files now generate an error instead of a warning since there
is no reason for having these empty files.
If a follow-up line in a Makefile is completely empty, there is no note
about trailing whitespace anymore since the warning about the misleading
empty line already covers this case.
The remaining code changes are only refactorings.
Changes since 0.4:
The log output in verbose mode surrounds each test now and says
immediately whether that test fails. Before, that information was
collected and only shown at the very end of the output, which was
inconvenient.
Changes since 19.3.18:
Small improvements to edge cases in SUBST blocks.
Small improvements to edge cases in variable assignment and alignment
of the continuation backslashes.
The --source option shows the changes from autofix, even when the
--show-autofix option is not given. This is a welcome side-effect of
making the autofix logging simpler and more predictable.
Changes since 19.3.17:
The SUBST check has been completely rewritten. It can handle several
SUBST classes at the same time now. This reduces the number of wrong
warnings.
Changes since 19.3.16:
Pkglint now handles SUBST blocks correctly, even those in which some of
the variables are defined conditionally. It correctly reports those that
are missing in at least one of the possible branches.
PKG_JVM is no longer marked as deprecated. It was once package-settable.
Since 2002 it is system-provided, and the package-settable counterpart
is PKG_JVM_DEFAULT. This does not fit into pkglint's simple model of
deprecating variables since the variable name is still valid, it should
just not be defined by packages anymore.
The alignment of variable assignments has been fixed in some edge cases.
In continuation lines where the backslash is beyond column 72, the
whitespace before the continuation backslash is fixed to a single space.
## Version 0.14.0 (2019-12-12)
* Ensure reverse dependencies are considered when a "pkgin install" triggers
an upgrade of a local package, avoiding potentially nasty bugs.
* Minor performance improvement.
Changes since 19.3.15:
When a package-settable variable gets a default value using the ?=
operator, pkglint no longer suggests to include bsd.prefs.mk, since that
doesn't make sense. Including bsd.prefs.mk only defines user-settable
and system-provided variables.
User and group names may be a single character only. While not widely
used, it's syntactically valid and there's no reason to prevent this.
In variable assignments, when pkglint removes unnecessary whitespace
between the variable name and the operator, it keeps the indentation of
the variable value the same as before. Previously, the indentation had
been changed, which required another run of pkglint --autofix.
PREFIX can only be used as a replacement for LOCALBASE after the whole
package Makefile has been loaded. This is because PREFIX is defined
very late, by bsd.pkg.mk. Therefore, don't suggest to replace LOCALBASE
with PREFIX in .if conditions.
When pkglint suggests to replace INSTALL_DATA_DIR commands with setting
INSTALLATION_DIRS instead, paths with a trailing slash are correctly
looked up in the PLIST. This suggests to use AUTO_MKDIRS more often.
Changes since 19.3.14:
Invalid lines in PLIST files are now reported as errors instead of
warnings. If pkglint doesn't know about it, it must be an error.
In PLIST files, all paths are validated to be canonical. That is, no
dotdot components, no absolute paths, no extra slashes, no intermediate
dot components.
Fewer notes for unexpanded variable expressions in DESCR files. Before,
the text $@ was reported as possible Makefile variable even though it
was just a Perl expression.
README files are allowed again in pkgsrc package directories. There was
no convincing argument why these should be forbidden.
A few diagnostics have been changed from NOTE to WARNING or from WARNING
to ERROR, to match their wording.
When pkglint suggests to replace :M with ==, the wording is now "can be
made" instead of "should".
Changes since 19.3.13:
When pkglint suggests to replace !empty(VARNAME:Mfixed) with ${VARNAME}
== fixed, the exact suggested expression is now part of the diagnostic.
The check and the autofix have been improved. They now apply only to the
last modifier in the whole chain, everything else was a bug in pkglint.
Pkglint now knows the scope of variables better than before. It knows
the difference between variables from <sys.mk> like MACHINE_ARCH, which
are always in scope, and those from mk/defaults/mk.conf, which only come
into scope later, after bsd.prefs.mk has been included. It warns when
variables are used too early, for example in .if conditions.
The pathnames in ALTERNATIVES files are now checked for absolute
pathnames. This mistake doesn't happen in practice, but the code for
converting the different path types internally made it necessary to add
these checks. At least this prevents typos.
The special check for obsolete licenses has been removed since their
license files have been removed and that is checked as well.
Variables named *_AWK may be appended to.
The variables _PKG_SILENT and _PKG_DEBUG are no longer deprecated, they
are obsolete now. They are not used in main pkgsrc and pkgsrc-wip
anymore.
When a package sets a default value for a user-settable variable (which
is something that should not happen anyway), it should .include
bsd.prefs.mk before, in order to not accidentally overwrite the
user-specified value.
Variable modifiers of the form :from=to are now parsed like in bmake.
They are greedy and eat up any following colons as well. This means that
${VAR:.c=.o:Q} replaces source.c with source.o:Q, instead of quoting it.
Pkglint now warns about such cases.
The handling of relative paths in diagnostics is now consistent. All
paths that are part of a diagnostic are relative to the line that issues
the diagnostic.
Fatal errors are no longer suppressed in --autofix mode.
Plus lots of refactoring, to prevent accidental mixing of incompatible
relative paths.
Changes since 19.3.12:
The command line option -Wspace has been removed. Warnings and notes
about whitespace are now generated by default and cannot be switched
off. This is to ensure a consistent visual appearance of the package
Makefiles.
Shell programs that are indented unnecessarily deep generate a note by
default now. Before, the option -Wall was necessary to get these notes.
The check for unintended comments in multi-line shell programs is now
enabled again. It had been disabled some time ago as byproduct of a bug
fix in the shell parser.
The check for unique buildlink3 package identifiers now also works if
pkglint is run from a package directory instead of the pkgsrc root
directory.
Changes since 19.3.11:
The command line option -Wstyle has been removed since it didn't have
any effect.
License names may contain underscores. This fixes 3 parse errors and 2
wrong notes about seemingly unused licenses.
The parser for Makefile variables has been improved for some edge cases.
The :M and :N modifiers behave surprisingly when they contain unbalanced
parentheses or braces. Pkglint now parses them in the same way as bmake,
but doesn't warn since these cases are not actually used in pkgsrc.
Changes since 19.3.10:
The check for buildlink3.mk files that are included conditionally in one
place and unconditionally in another place have been refined. Now they
also work in cases that do not involve any variables, such as when the
condition is a mere exists(filename).
References to variables that use parentheses instead of the usual braces
produce a warning now, even if pkglint cannot fix them automatically.
This affects only a few instances where more than one such variable
reference appeared in a single line.
The --log-verbose command line option has been removed since it does not
have any practical use other than improving the performance during
pkglint development itself. Because of that it hadn't even been
mentioned in the manual page.
Warnings for missing license files now report the path to the license
file relative to the line where the warning occurs, like everywhere
else.
Changes since 19.3.9:
In diagnostics for suggested package updates, the exact line of doc/TODO
is mentioned. If a suggested update has an additional comment, the
brackets around that comment are not output anymore.
The check for defined but not used variables has been improved for the
edge case of defining a variable in the package Makefile and using it
in the buildlink3.mk file of the same package, which just doesn't work.
Makefile fragments in patches/ directories are now completely ignored.
It was a hypothetical case anyway.
Comparing PKGSRC_COMPILER using the == or != operators is now considered
an error instead of a warning. The common cases can be autofixed.
Changes since 19.3.8:
Match man pages in ALTERNATIVES with their counterparts in PLIST. In
PLIST files, ${PKGMANDIR} may be abbreviated as a simple "man", but not
in ALTERNATIVES.
Changes since 4.193.1:
Disabled unreliable check about possible typos in variable names. For
example, in pkgtools/url2pkg the variable TEST_DEPENDS was flagged even
though it is well-known.
Disabled unreliable check about unknown shell commands. For example, in
pkgtools/pkglint4 all commands from ${PREFIX} are ok, but were still
flagged.
Changes since 19.3.6:
Improved variable value alignment.
Fixed wrong warning about comment lines that were interpreted as shell
commands before.
Warn when the first category of a package doesn't correspond to the
path in the filesystem. This affects 603 packages.
No longer warn about deprecated BUILDLINK_TRANSFORM.${OPSYS}. The
deprecation warning was meant for BUILDLINK_TRANSFORM.${pkgbase}, but
since pkglint cannot distinguish between these, the warnings were wrong.
Some OS (e.g. Arch Linux) do not have hostname in their base set of
packages.
pkgsrc changes:
---------------
* Replace $(hostname) by $(uname -n).
Thanks jperkin@!
Changes since 19.3.5:
Improved indentation and alignment of multi-line variable assignments.
Improved indentation of multi-line shell commands.
Added warning for adding unquoted words to PKG_FAIL_REASON, which is a
list of messages, one per line.
Lines that start with tabs followed by a # are not shell commands, they
are comments. Bmake treats them in the same way.
Change the type of BROKEN to be a list of messages, instead of a single
message. This allows at least a bit of formatting in the error messages.
Changes since 19.3.4:
Variable uses in parentheses (such as $(VAR) instead of ${VAR}) are
treated the same. The ones in parentheses had less support before.
Improved the checks for options.mk files, adding support for options
that are defined using .for loops and those referring to other
variables.
Packages that set DISTFILES to an empty list no longer require a
distinfo file.
Patches whose filename contains the word CVE may patch more than one
target file.
Changes since 19.3.3:
In cases where the conditions for including buildlink3.mk files differ
between the package itself and its own buildlink3.mk file, explain how
to determine PKG_OPTIONS for dependencies.
Don't issue wrong warnings in options.mk files when the options are
handled in a .for loop.
Changes since 19.3.3:
The code that handles license names is the same for Python and Perl
modules.
Python packages that use distutils.extension can be handled. An example
package is devel/py-pysha3.
The DESCR file is filled from the package's README file.
The PLIST is generated with some comments providing further assistance.
Changes since 19.3.2:
The rationale for variables like BROKEN, GCC_REQD and for direct
inclusion of builtin.mk files may span multiple lines, and it may end
with an empty comment line.
Changes since 19.3.1:
* Pkglint no longer warns about a missing :Q modifier if there is also
a :D modifier, since the latter hides the original variable value
from the expression value.
* Variable names like .CURDIR are now allowed in the _VARGROUPS section.
* In dependency lines like "${_COOKIE.extract}:", pkglint no longer
warns about the unknown target. No matter whether this is a file name
or even a list of other targets, there's no chance for a typo here.
* If some dependencies are included conditionally, and the package
Makefile and buildlink3.mk disagree, and the conditions depend on
PKG_OPTIONS, pkglint outputs a helpful explanation.
* The check for including builtin.mk directly can be disabled by giving
a reason in a comment at the end of the line.
Placing the code into a main function allows the tests to source the file
without any side effects.
The LC_ALL=C environment variable is necessary on NetBSD/x86_64 8.0 in
the en_US.UTF-8 locale since otherwise R dumps core in the source(...)
call, deep inside towctrans.
Before, it was difficult to edit the R code since all dollars,
backslashes and backticks had to be escaped. Separating the code also
makes the R part testable on its own.
Changes since 0.6:
* Run properly even if TMPDIR is not set.
* Quote variables to prevent unintended interpolation.
* Write error messages to stderr instead of stdout.
Changes since 19.3.0:
* In buildlink3.mk files, the package identifier doesn't have to match
the PKGBASE from the package Makefile exactly. The PKGBASE may have a
leading "lib" (for libiconv and libgettext), as well as a trailing
number (for emacs20 and netatalk22).
* GITHUB_RELEASE is added to the variables that should appear in a fixed
order in the package Makefile.
* In the MASTER_SITE URLs, the transport protocol is irrelevant for
matching direct URLs to the predefined MASTER_SITE_* variables.
The motivation is to consider meson as an application, so there is only
one copy on the system, and as a python 3 program, it can build python 2
packages.
Changes since 19.3.2:
* Replaced the Perl implementation with the Python implementation.
* Improved detection of WRKSRC: only directories are candidates, regular
files are ignored.
* R packages are handled by R2pkg instead of url2pkg.
The Python implementation has received more bugfixes, new features and
handles all kinds of edge cases better. All previous features have been
either copied or intentionally omitted.
CFLAGS and LDFLAGS had been checked for GCC-style flags only. This didn't
make sense since wrong flags would be detected on the exotic platforms
anyway.
The experimental --autofix has been removed.
The check for "# used by" comments in Makefile.common has been removed.
Especially the --autofix could damage files.
The warnings for ICON_THEMES have been removed since they appear in
groups of thousands, and the old pkglint cannot issue warnings "once per
file" simply.
The warnings about variable permissions were not understandable enough to
be acted upon. The new pkglint does this better.
The languages that are allowed in USE_LANGUAGES are defined differently
in the pkgsrc infrastructure, thus the old parsing code does not work
anymore. Therefore all identifiers are allowed now.
Dependency patterns like 'package>=1.0<2.0' are no longer marked as
wrong.
The debatable warning about plural names is gone.
The order of variables in simple Makefiles is no longer checked. Some new
variables have been added in the meantime, and keeping the consistent
order is not of utmost importance to those pkgsrc developers who work
with pkglint4. They are experienced enough to know the rules.
Missing manual pages are no longer marked in the PLIST files. It's not
the job of pkgsrc to provide these files.
The warning about unnoticed errors in pipelines like 'find | xargs' has
been removed because the shell parser is unreliable. This is solved
better in the new pkglint.
The version number '4.193.0' had always been compared to the string
'pkglint4-4.193.0', which when interpreted as a version number, was:
(16, 11, 7, 12, 9, 14, 20, 4, 0, 193, 0, 0). This was always greater than
the (4, 0, 193, 0, 0).
Changes since 19.3.1 (only to the Python implementation):
* In Python packages, the variables LICENSE, HOMEPAGE and COMMENT are
filled from their counterparts in setup.py, no matter whether
the package uses setuptools or distutils.core.
* For buildlink3 dependencies, the variables BUILDLINK_DEPENDS and
BUILDLINK_API_DEPENDS are added to the package Makefile.
* The DESCR and PLIST files are only created if they are missing.
Existing files are not overwritten.
* Packages that contain .po files will set USE_PKGLOCALEDIR=yes.
Before, only packages that contained .mo or .gmo did that.
* More download URLs from SourceForge can be handled.
* SourceForge packages get their HOMEPAGE set correctly.
* When matching the URL against the MASTER_SITE_* variables, the
protocol is ignored. This allows https URLs to use the
MASTER_SITE_GNU, which still uses http.
* Python packages from GitHub only get their MASTER_SITES converted to
PyPI if they can be actually downloaded from there as well. This
check was missing before.
The MASTER_SITE_GNU still uses http:// for downloading the distfiles.
Since typical URLs today use https://, ignore the protocol when matching
the given URL against the MASTER_SITE_* variables from mk/fetch/sites.mk.
The previous version of the code was largely work in progress. Now the code
has been grouped and sorted. A few bugs have been fixed on the way:
* If a PKGNAME had been added in the interactive editor session, it had
been overwritten before. This was because of a typo.
* The whole code has been grouped into classes, to clearly show the
dependencies between the parts.
* Generation of the initial Makefile has been split into smaller methods,
to make them individually testable and to reduce the scope of the local
variables.
* When creating a package in a directory pkgsrc/local/*, "local" is not
used as the primary category of the package.
* GNU configure and other configure scripts are also detected if they are
not placed directly in WRKSRC.
* Packages that contain *.po files will have USE_PKGLOCALEDIR=yes in the
package Makefile. Previously, only *.mo or *.gmo files triggered this
variable.
* When PKGNAME is based on DISTNAME, it is only written to the package
Makefile if there is an actual prefix or transformation.
The files DESCR and PLIST are no longer overwritten. They are only
created if they don't exist.
The GConf schemas detection had written the include line too often, once
for every schema file.
Changes since 19.3.0:
* Added an alternative url2pkg implementation in Python that will
eventually replace the Perl implementation. Reasons are:
* Perl is cumbersome to type with all the special characters
* Perl does not even have a Boolean type, after all the years
* Perl cannot check the number of arguments passed to subs
* Python does not suffer from the above limitations
* Python is available on as many platforms as Perl
* The two implementations will be kept in sync until all the features
have made it into the Python implementation, and everything has
automatic tests.
* Added the -v or --verbose command line option.
Changes since 5.7.24:
* There is no need to ask the dummy MAINTAINER from url2pkg whether
committing changes is ok.
* When autofixing a condition like !empty(PKGPATH:Mliteral), don't
generate unnecessary parentheses around ${PKGPATH} == literal.
* In a _VARGROUPS section, the public variables should be listed
before the private variables, to put important things first.
* When pkglint suggests to be run again with the -e, -fs or -F options,
repeat the whole command line, to allow for copy-and-paste.
* The checks for PKGPATH are fixed and enhanced. It is not a relative
path like in ../../category/package, but relative to the pkgsrc root.
* Unintended file globbing in sed commands such as s,.*,any, gets a
warning.
* MASTER_SITES should normally end with a slash, in rare cases an
equals sign or a colon are correct, too.
* Detect redundant directories in INSTALLATION_DIRS.
Changes since 2.37:
* Test dependencies in Perl MakeMaker modules are copied to the
package Makefile.
* Unknown dependencies in Perl MakeMaker modules no longer abort url2pkg
but instead are added as "# TODO" dependencies to the package Makefile.
* The license from Perl MakeMaker and Python modules is copied to the
package Makefile.
* The build dependencies, COMMENT and HOMEPAGE from Python setuptools
modules are copied to the package Makefile.
* No more "undefined $url" when running url2pkg in a directory that
already has a package Makefile and is extracted.
Changes since 2.36:
* Fixed generation of the package Makefile. Due to a typo, the line that
included bsd.pkg.mk had been omitted, making any generated package fail.
## Version 0.13.0 (2019-09-24)
* "pkgin upgrade" now upgrades all packages instead of only keep packages, and
essentially replaces "pkgin full-upgrade". Selective upgrades have always
been risky and provide no benefit, especially in light of refresh support.
"pkgin full-upgrade" is retained for compatibility but marked as deprecated.
* Refresh the local package repository when a forced update is requested, so
that a failed update can be repaired easily.
* Switch NetBSD default repository from ftp.netbsd.org to cdn.netbsd.org.
Changes since 5.7.23:
* Improved the _VARGROUPS check for ignored variables
* Removed wrong warnings for variable expressions like ${VAR:Dyes:Uno}
* Used correct terminology for the :Q modifier (it's not an operator)
Changes since 2.33:
* code cleanup in MakeMaker.pm
* don't create a PLIST for Perl packages
* allow Python dependencies from setup.py to use spaces around >=
* properly indented Python code according to PEP 8
* added mock for setuptools.find_package
* only try to migrate a package from GitHub to PyPI if it is really
a Python package on GitHub
Changes since 5.7.22:
* Added a warning for lines that look empty but are actually follow-up
lines from a previous line.
* Added notes for unusual placement of the continuation backslash. It
should always be preceded by a single space or tab, or be in column
73.
* Improved check for needlessly complicated !empty(PKGPATH:Mcat/pkg)
that is transformed into the simpler ${PKGPATH} == cat/pkg, even if
the package name contains hyphens, dots, plus or slashes.
* Added check for the _VARGROUPS section since that section contains
many redundancies that can easily be checked. For example, in
mk/compiler/gcc.mk the _VARGROUPS section is 67 lines long and
contains far more than 100 variables. It's tedious to manually check
this file for internal consistency. That's better left to pkglint.
* The empty variable is no longer flagged as "used but not defined".
It appears in expressions like ${:Ustring}.
* When --source is combined with --explain, don't unnecessarily
repeat the source code for a single line if there are several
diagnostics. Instead, even omit the empty line between the diagnostics
for the same line.
Start building FreeGLUT with GLUT compatibility on.
- GLUT's maintainance state is much worse than FreeGLUT's, with no
substanital commits since 2011.
- GLUT can no longer be consider a core component of Mesa or X11.
- Nobody except us seems to be using it...
Other vendors still shipping GLUT:
- GoboLinux, T2 SDE, PLD Linux
Revbump packages using GLUT or freeglut by default.
Be more tolerant of not finding PKGPATH, and basically attempt to skip
the package rather than treating it as a fatal error. (This was a
stray edit that I've been running with for a while.)
Changes since 5.7.21:
* The files from wip/mk do not belong to the main pkgsrc infrastructure.
Therefore, when loading the package Makefile to look for defined but
unused variables, parsing doesn't stop in these files.
Changes since 5.7.21:
* The files from wip/mk do not belong to the main pkgsrc infrastructure.
Therefore, when loading the package Makefile to look for defined but
unused variables, parsing doesn't stop in these files.
Changes since 5.7.20:
* PKG_OPTIONS that are handled using patterns are correctly identified.
* Simple R packages should follow the canonical variable order.
* Fixed some edge cases for aligning variable assignments.
* Improved detection of allowed values for USE_LANGUAGES.
