9 QA Error and Warning Messages

9.1 Introduction

When building a recipe, the OpenEmbedded build system performs various QA checks on the output to ensure that common issues are detected and reported. Sometimes when you create a new recipe to build new software, it will build with no problems. When this is not the case, or when you have QA issues building any software, it could take a little time to resolve them.

While it is tempting to ignore a QA message or even to disable QA checks, it is best to try and resolve any reported QA issues. This chapter provides a list of the QA messages and brief explanations of the issues you could encounter so that you can properly resolve problems.

The next section provides a list of all QA error and warning messages based on a default configuration. Each entry provides the message or error form along with an explanation.

Note

At the end of each message, the name of the associated QA test (as listed in the “insane.bbclass” section) appears within square brackets.
As mentioned, this list of error and warning messages is for QA checks only. The list does not cover all possible build errors or warnings you could encounter.
Because some QA checks are disabled by default, this list does not include all possible QA check errors and warnings.

9.2 Errors and Warnings

<packagename>: <path> is using libexec please relocate to <libexecdir> [libexec]

The specified package contains files in /usr/libexec when the distro configuration uses a different path for <libexecdir> By default, <libexecdir> is $prefix/libexec. However, this default can be changed (e.g. ${libdir}).

package <packagename> contains bad RPATH <rpath> in file <file> [rpaths]

The specified binary produced by the recipe contains dynamic library load paths (rpaths) that contain build system paths such as TMPDIR, which are incorrect for the target and could potentially be a security issue. Check for bad -rpath options being passed to the linker in your do_compile log. Depending on the build system used by the software being built, there might be a configure option to disable rpath usage completely within the build of the software.

<packagename>: <file> contains probably-redundant RPATH <rpath> [useless-rpaths]

The specified binary produced by the recipe contains dynamic library load paths (rpaths) that on a standard system are searched by default by the linker (e.g. /lib and /usr/lib). While these paths will not cause any breakage, they do waste space and are unnecessary. Depending on the build system used by the software being built, there might be a configure option to disable rpath usage completely within the build of the software.

<packagename> requires <files>, but no providers in its RDEPENDS [file-rdeps]

A file-level dependency has been identified from the specified package on the specified files, but there is no explicit corresponding entry in RDEPENDS. If particular files are required at runtime then RDEPENDS should be declared in the recipe to ensure the packages providing them are built.

<packagename1> rdepends on <packagename2>, but it isn't a build dependency? [build-deps]

There is a runtime dependency between the two specified packages, but there is nothing explicit within the recipe to enable the OpenEmbedded build system to ensure that dependency is satisfied. This condition is usually triggered by an RDEPENDS value being added at the packaging stage rather than up front, which is usually automatic based on the contents of the package. In most cases, you should change the recipe to add an explicit RDEPENDS for the dependency.

non -dev/-dbg/nativesdk- package contains symlink .so: <packagename> path '<path>' [dev-so]

Symlink .so files are for development only, and should therefore go into the -dev package. This situation might occur if you add *.so* rather than *.so.* to a non-dev package. Change FILES (and possibly PACKAGES) such that the specified .so file goes into an appropriate -dev package.

non -staticdev package contains static .a library: <packagename> path '<path>' [staticdev]

Static .a library files should go into a -staticdev package. Change FILES (and possibly PACKAGES) such that the specified .a file goes into an appropriate -staticdev package.

<packagename>: found library in wrong location [libdir]

The specified file may have been installed into an incorrect (possibly hardcoded) installation path. For example, this test will catch recipes that install /lib/bar.so when ${base_libdir} is “lib32”. Another example is when recipes install /usr/lib64/foo.so when ${libdir} is “/usr/lib”. False positives occasionally exist. For these cases add “libdir” to INSANE_SKIP for the package.

non debug package contains .debug directory: <packagename> path <path> [debug-files]

The specified package contains a .debug directory, which should not appear in anything but the -dbg package. This situation might occur if you add a path which contains a .debug directory and do not explicitly add the .debug directory to the -dbg package. If this is the case, add the .debug directory explicitly to FILES:${PN}-dbg. See FILES for additional information on FILES.

<packagename> installs files in <path>, but it is expected to be empty [empty-dirs]

The specified package is installing files into a directory that is normally expected to be empty (such as /tmp). These files may be more appropriately installed to a different location, or perhaps alternatively not installed at all, usually by updating the do_install task/function.

Architecture did not match (<file_arch>, expected <machine_arch>) in <file> [arch]

By default, the OpenEmbedded build system checks the Executable and Linkable Format (ELF) type, bit size, and endianness of any binaries to ensure they match the target architecture. This test fails if any binaries do not match the type since there would be an incompatibility. The test could indicate that the wrong compiler or compiler options have been used. Sometimes software, like bootloaders, might need to bypass this check. If the file you receive the error for is firmware that is not intended to be executed within the target operating system or is intended to run on a separate processor within the device, you can add “arch” to INSANE_SKIP for the package. Another option is to check the do_compile log and verify that the compiler options being used are correct.
Bit size did not match (<file_bits>, expected <machine_bits>) in <file> [arch]

By default, the OpenEmbedded build system checks the Executable and Linkable Format (ELF) type, bit size, and endianness of any binaries to ensure they match the target architecture. This test fails if any binaries do not match the type since there would be an incompatibility. The test could indicate that the wrong compiler or compiler options have been used. Sometimes software, like bootloaders, might need to bypass this check. If the file you receive the error for is firmware that is not intended to be executed within the target operating system or is intended to run on a separate processor within the device, you can add “arch” to INSANE_SKIP for the package. Another option is to check the do_compile log and verify that the compiler options being used are correct.
Endianness did not match (<file_endianness>, expected <machine_endianness>) in <file> [arch]

By default, the OpenEmbedded build system checks the Executable and Linkable Format (ELF) type, bit size, and endianness of any binaries to ensure they match the target architecture. This test fails if any binaries do not match the type since there would be an incompatibility. The test could indicate that the wrong compiler or compiler options have been used. Sometimes software, like bootloaders, might need to bypass this check. If the file you receive the error for is firmware that is not intended to be executed within the target operating system or is intended to run on a separate processor within the device, you can add “arch” to INSANE_SKIP for the package. Another option is to check the do_compile log and verify that the compiler options being used are correct.

ELF binary '<file>' has relocations in .text [textrel]

The specified ELF binary contains relocations in its .text sections. This situation can result in a performance impact at runtime.

Typically, the way to solve this performance issue is to add “-fPIC” or “-fpic” to the compiler command-line options. For example, given software that reads CFLAGS when you build it, you could add the following to your recipe:
```
CFLAGS:append = " -fPIC "
```
For more information on text relocations at runtime, see https://www.akkadia.org/drepper/textrelocs.html.

File '<file>' in package '<package>' doesn't have GNU_HASH (didn't pass LDFLAGS?) [ldflags]

This indicates that binaries produced when building the recipe have not been linked with the LDFLAGS options provided by the build system. Check to be sure that the LDFLAGS variable is being passed to the linker command. A common workaround for this situation is to pass in LDFLAGS using TARGET_CC_ARCH within the recipe as follows:
```
TARGET_CC_ARCH += "${LDFLAGS}"
```

Package <packagename> contains Xorg driver (<driver>) but no xorg-abi- dependencies [xorg-driver-abi]

The specified package contains an Xorg driver, but does not have a corresponding ABI package dependency. The xserver-xorg recipe provides driver ABI names. All drivers should depend on the ABI versions that they have been built against. Driver recipes that include xorg-driver-input.inc or xorg-driver-video.inc will automatically get these versions. Consequently, you should only need to explicitly add dependencies to binary driver recipes.

The /usr/share/info/dir file is not meant to be shipped in a particular package. [infodir]

The /usr/share/info/dir should not be packaged. Add the following line to your do_install task or to your do_install:append within the recipe as follows:
```
rm ${D}${infodir}/dir
```

Symlink <path> in <packagename> points to TMPDIR [symlink-to-sysroot]

The specified symlink points into TMPDIR on the host. Such symlinks will work on the host. However, they are clearly invalid when running on the target. You should either correct the symlink to use a relative path or remove the symlink.

<file> failed sanity test (workdir) in path <path> [la]

The specified .la file contains TMPDIR paths. Any .la file containing these paths is incorrect since libtool adds the correct sysroot prefix when using the files automatically itself.

<file> failed sanity test (tmpdir) in path <path> [pkgconfig]

The specified .pc file contains TMPDIR/WORKDIR paths. Any .pc file containing these paths is incorrect since pkg-config itself adds the correct sysroot prefix when the files are accessed.

<packagename> rdepends on <debug_packagename> [debug-deps]

There is a dependency between the specified non-dbg package (i.e. a package whose name does not end in -dbg) and a package that is a dbg package. The dbg packages contain debug symbols and are brought in using several different methods:
- Using the dbg-pkgs IMAGE_FEATURES value.
- Using IMAGE_INSTALL.
- As a dependency of another dbg package that was brought in using one of the above methods.
The dependency might have been automatically added because the dbg package erroneously contains files that it should not contain (e.g. a non-symlink .so file) or it might have been added manually (e.g. by adding to RDEPENDS).

<packagename> rdepends on <dev_packagename> [dev-deps]

There is a dependency between the specified non-dev package (a package whose name does not end in -dev) and a package that is a dev package. The dev packages contain development headers and are usually brought in using several different methods:
- Using the dev-pkgs IMAGE_FEATURES value.
- Using IMAGE_INSTALL.
- As a dependency of another dev package that was brought in using one of the above methods.
The dependency might have been automatically added (because the dev package erroneously contains files that it should not have (e.g. a non-symlink .so file) or it might have been added manually (e.g. by adding to RDEPENDS).

<var>:<packagename> is invalid: <comparison> (<value>) only comparisons <, =, >, <=, and >= are allowed [dep-cmp]

If you are adding a versioned dependency relationship to one of the dependency variables (RDEPENDS, RRECOMMENDS, RSUGGESTS, RPROVIDES, RREPLACES, or RCONFLICTS), you must only use the named comparison operators. Change the versioned dependency values you are adding to match those listed in the message.

<recipename>: The compile log indicates that host include and/or library paths were used. Please check the log '<logfile>' for more information. [compile-host-path]

The log for the do_compile task indicates that paths on the host were searched for files, which is not appropriate when cross-compiling. Look for “is unsafe for cross-compilation” or “CROSS COMPILE Badness” in the specified log file.

<recipename>: The install log indicates that host include and/or library paths were used. Please check the log '<logfile>' for more information. [install-host-path]

The log for the do_install task indicates that paths on the host were searched for files, which is not appropriate when cross-compiling. Look for “is unsafe for cross-compilation” or “CROSS COMPILE Badness” in the specified log file.

This autoconf log indicates errors, it looked at host include and/or library paths while determining system capabilities. Rerun configure task after fixing this. [configure-unsafe]

The log for the do_configure task indicates that paths on the host were searched for files, which is not appropriate when cross-compiling. Look for “is unsafe for cross-compilation” or “CROSS COMPILE Badness” in the specified log file.

<packagename> doesn't match the [a-z0-9.+-]+ regex [pkgname]

The convention within the OpenEmbedded build system (sometimes enforced by the package manager itself) is to require that package names are all lower case and to allow a restricted set of characters. If your recipe name does not match this, or you add packages to PACKAGES that do not conform to the convention, then you will receive this error. Rename your recipe. Or, if you have added a non-conforming package name to PACKAGES, change the package name appropriately.

<recipe>: configure was passed unrecognized options: <options> [unknown-configure-option]

The configure script is reporting that the specified options are unrecognized. This situation could be because the options were previously valid but have been removed from the configure script. Or, there was a mistake when the options were added and there is another option that should be used instead. If you are unsure, consult the upstream build documentation, the ./configure --help output, and the upstream change log or release notes. Once you have worked out what the appropriate change is, you can update EXTRA_OECONF, PACKAGECONFIG_CONFARGS, or the individual PACKAGECONFIG option values accordingly.

Recipe <recipefile> has PN of "<recipename>" which is in OVERRIDES, this can result in unexpected behavior. [pn-overrides]

The specified recipe has a name (PN) value that appears in OVERRIDES. If a recipe is named such that its PN value matches something already in OVERRIDES (e.g. PN happens to be the same as MACHINE or DISTRO), it can have unexpected consequences. For example, assignments such as FILES:${PN} = "xyz" effectively turn into FILES = "xyz". Rename your recipe (or if PN is being set explicitly, change the PN value) so that the conflict does not occur. See FILES for additional information.

<recipefile>: Variable <variable> is set as not being package specific, please fix this. [pkgvarcheck]

Certain variables (RDEPENDS, RRECOMMENDS, RSUGGESTS, RCONFLICTS, RPROVIDES, RREPLACES, FILES, pkg_preinst, pkg_postinst, pkg_prerm, pkg_postrm, and ALLOW_EMPTY) should always be set specific to a package (i.e. they should be set with a package name override such as RDEPENDS:${PN} = "value" rather than RDEPENDS = "value"). If you receive this error, correct any assignments to these variables within your recipe.
recipe uses DEPENDS:${PN}, should use DEPENDS [pkgvarcheck]

This check looks for instances of setting DEPENDS:${PN} which is erroneous (DEPENDS is a recipe-wide variable and thus it is not correct to specify it for a particular package, nor will such an assignment actually work.) Set DEPENDS instead.

File '<file>' from <recipename> was already stripped, this will prevent future debugging! [already-stripped]

Produced binaries have already been stripped prior to the build system extracting debug symbols. It is common for upstream software projects to default to stripping debug symbols for output binaries. In order for debugging to work on the target using -dbg packages, this stripping must be disabled.

Depending on the build system used by the software being built, disabling this stripping could be as easy as specifying an additional configure option. If not, disabling stripping might involve patching the build scripts. In the latter case, look for references to “strip” or “STRIP”, or the “-s” or “-S” command-line options being specified on the linker command line (possibly through the compiler command line if preceded with “-Wl,”).

Note

Disabling stripping here does not mean that the final packaged binaries will be unstripped. Once the OpenEmbedded build system splits out debug symbols to the -dbg package, it will then strip the symbols from the binaries.

<packagename> is listed in PACKAGES multiple times, this leads to packaging errors. [packages-list]

Package names must appear only once in the PACKAGES variable. You might receive this error if you are attempting to add a package to PACKAGES that is already in the variable’s value.

FILES variable for package <packagename> contains '//' which is invalid. Attempting to fix this but you should correct the metadata. [files-invalid]

The string “//” is invalid in a Unix path. Correct all occurrences where this string appears in a FILES variable so that there is only a single “/”.

<recipename>: Files/directories were installed but not shipped in any package [installed-vs-shipped]

Files have been installed within the do_install task but have not been included in any package by way of the FILES variable. Files that do not appear in any package cannot be present in an image later on in the build process. You need to do one of the following:
- Add the files to FILES for the package you want them to appear in (e.g. FILES:${PN} for the main package).
- Delete the files at the end of the do_install task if the files are not needed in any package.
<oldpackage>-<oldpkgversion> was registered as shlib provider for <library>, changing it to <newpackage>-<newpkgversion> because it was built later

This message means that both <oldpackage> and <newpackage> provide the specified shared library. You can expect this message when a recipe has been renamed. However, if that is not the case, the message might indicate that a private version of a library is being erroneously picked up as the provider for a common library. If that is the case, you should add the library’s .so filename to PRIVATE_LIBS in the recipe that provides the private version of the library.

LICENSE:<packagename> includes licenses (<licenses>) that are not listed in LICENSE [unlisted-pkg-lics]

The LICENSE of the recipe should be a superset of all the licenses of all packages produced by this recipe. In other words, any license in LICENSE:* should also appear in LICENSE.

AM_GNU_GETTEXT used but no inherit gettext [configure-gettext]

If a recipe is building something that uses automake and the automake files contain an AM_GNU_GETTEXT directive then this check will fail if there is no inherit gettext statement in the recipe to ensure that gettext is available during the build. Add inherit gettext to remove the warning.

package contains mime types but does not inherit mime: <packagename> path '<file>' [mime]

The specified package contains mime type files (.xml files in ${datadir}/mime/packages) and yet does not inherit the mime class which will ensure that these get properly installed. Either add inherit mime to the recipe or remove the files at the do_install step if they are not needed.

package contains desktop file with key 'MimeType' but does not inhert mime-xdg: <packagename> path '<file>' [mime-xdg]

The specified package contains a .desktop file with a ‘MimeType’ key present, but does not inherit the mime-xdg class that is required in order for that to be activated. Either add inherit mime to the recipe or remove the files at the do_install step if they are not needed.

<recipename>: SRC_URI uses unstable GitHub archives [src-uri-bad]

GitHub provides “archive” tarballs, however these can be re-generated on the fly and thus the file’s signature will not necessarily match that in the SRC_URI checksums in future leading to build failures. It is recommended that you use an official release tarball or switch to pulling the corresponding revision in the actual git repository instead.
SRC_URI uses PN not BPN [src-uri-bad]

If some part of SRC_URI needs to reference the recipe name, it should do so using ${BPN} rather than ${PN} as the latter will change for different variants of the same recipe e.g. when BBCLASSEXTEND or multilib are being used. This check will fail if a reference to ${PN} is found within the SRC_URI value - change it to ${BPN} instead.

<recipename>: recipe doesn't inherit features_check [unhandled-features-check]

This check ensures that if one of the variables that the features_check class supports (e.g. REQUIRED_DISTRO_FEATURES) is used, then the recipe inherits features_check in order for the requirement to actually work. If you are seeing this message, either add inherit features_check to your recipe or remove the reference to the variable if it is not needed.

<recipename>: recipe defines ALTERNATIVE:<packagename> but doesn't inherit update-alternatives. This might fail during do_rootfs later! [missing-update-alternatives]

This check ensures that if a recipe sets the ALTERNATIVE variable that the recipe also inherits update-alternatives such that the alternative will be correctly set up. If you are seeing this message, either add inherit update-alternatives to your recipe or remove the reference to the variable if it is not needed.

<packagename>: <file> maximum shebang size exceeded, the maximum size is 128. [shebang-size]

This check ensures that the shebang line (#! in the first line) for a script is not longer than 128 characters, which can cause an error at runtime depending on the operating system. If you are seeing this message then the specified script may need to be patched to have a shorter in order to avoid runtime problems.

<packagename> contains perllocal.pod (<files>), should not be installed [perllocalpod]

perllocal.pod is an index file of locally installed modules and so shouldn’t be installed by any distribution packages. The cpan class already sets NO_PERLLOCAL to stop this file being generated by most Perl recipes, but if a recipe is using MakeMaker directly then they might not be doing this correctly. This check ensures that perllocal.pod is not in any package in order to avoid multiple packages shipping this file and thus their packages conflicting if installed together.

<packagename> package is not obeying usrmerge distro feature. /<path> should be relocated to /usr. [usrmerge]

If usrmerge is in DISTRO_FEATURES, this check will ensure that no package installs files to root (/bin, /sbin, /lib, /lib64) directories. If you are seeing this message, it indicates that the do_install step (or perhaps the build process that do_install is calling into, e.g. make install is using hardcoded paths instead of the variables set up for this (bindir, sbindir, etc.), and should be changed so that it does.

Fuzz detected: <patch output> [patch-fuzz]
This check looks for evidence of “fuzz” when applying patches within the do_patch task. Patch fuzz is a situation when the patch tool ignores some of the context lines in order to apply the patch. Consider this example:

Patch to be applied:
--- filename +++ filename context line 1 context line 2 context line 3 +newly added line context line 4 context line 5 context line 6
Original source code:
different context line 1 different context line 2 context line 3 context line 4 different context line 5 different context line 6
Outcome (after applying patch with fuzz):
different context line 1 different context line 2 context line 3 newly added line context line 4 different context line 5 different context line 6
Chances are, the newly added line was actually added in a completely wrong location, or it was already in the original source and was added for the second time. This is especially possible if the context line 3 and 4 are blank or have only generic things in them, such as #endif or }. Depending on the patched code, it is entirely possible for an incorrectly patched file to still compile without errors.

How to eliminate patch fuzz warnings

Use the devtool command as explained by the warning. First, unpack the source into devtool workspace:
devtool modify <recipe>
This will apply all of the patches, and create new commits out of them in the workspace - with the patch context updated.

Then, replace the patches in the recipe layer:
devtool finish --force-patch-refresh <recipe> <layer_path>
The patch updates then need be reviewed (preferably with a side-by-side diff tool) to ensure they are indeed doing the right thing i.e.:
1. they are applied in the correct location within the file;
2. they do not introduce duplicate lines, or otherwise do things that are no longer necessary.
To confirm these things, you can also review the patched source code in devtool’s workspace, typically in <build_dir>/workspace/sources/<recipe>/

Once the review is done, you can create and publish a layer commit with the patch updates that modify the context. Devtool may also refresh other things in the patches, those can be discarded.

9.3 Configuring and Disabling QA Checks

You can configure the QA checks globally so that specific check failures either raise a warning or an error message, using the WARN_QA and ERROR_QA variables, respectively. You can also disable checks within a particular recipe using INSANE_SKIP. For information on how to work with the QA checks, see the “insane.bbclass” section.

Note

Please keep in mind that the QA checks are meant to detect real or potential problems in the packaged output. So exercise caution when disabling these checks.