1dpkg-gensymbols(1) dpkg suite dpkg-gensymbols(1)
2
6 dpkg-gensymbols - generate symbols files (shared library dependency
7 information)
8
10 dpkg-gensymbols [option...]
11
13 dpkg-gensymbols scans a temporary build tree (debian/tmp by default)
14 looking for libraries and generates a symbols file describing them.
15 This file, if non-empty, is then installed in the DEBIAN subdirectory
16 of the build tree so that it ends up included in the control
17 information of the package.
18 When generating those files, it uses as input some symbols files
20 provided by the maintainer. It looks for the following files (and uses
21 the first that is found):
22 · debian/package.symbols.arch
24 · debian/symbols.arch
26 · debian/package.symbols
28 · debian/symbols
30 The main interest of those files is to provide the minimal version
32 associated to each symbol provided by the libraries. Usually it
33 corresponds to the first version of that package that provided the
34 symbol, but it can be manually incremented by the maintainer if the ABI
35 of the symbol is extended without breaking backwards compatibility.
36 It's the responsibility of the maintainer to keep those files up-to-
37 date and accurate, but dpkg-gensymbols helps with that.
38 When the generated symbols files differ from the maintainer supplied
40 one, dpkg-gensymbols will print a diff between the two versions.
41 Furthermore if the difference is too significant, it will even fail
42 (you can customize how much difference you can tolerate, see the -c
43 option).
44
46 The symbols files are really useful only if they reflect the evolution
47 of the package through several releases. Thus the maintainer has to
48 update them every time that a new symbol is added so that its
49 associated minimal version matches reality. The diffs contained in the
50 build logs can be used as a starting point, but the maintainer,
51 additionally, has to make sure that the behaviour of those symbols has
52 not changed in a way that would make anything using those symbols and
53 linking against the new version, stop working with the old version. In
54 most cases, the diff applies directly to the debian/package.symbols
55 file. That said, further tweaks are usually needed: it's recommended
56 for example to drop the Debian revision from the minimal version so
57 that backports with a lower version number but the same upstream
58 version still satisfy the generated dependencies. If the Debian
59 revision can't be dropped because the symbol really got added by the
60 Debian specific change, then one should suffix the version with ‘~’.
61 Before applying any patch to the symbols file, the maintainer should
63 double-check that it's sane. Public symbols are not supposed to
64 disappear, so the patch should ideally only add new lines.
65 Note that you can put comments in symbols files: any line with ‘#’ as
67 the first character is a comment except if it starts with ‘#include’
68 (see section Using includes). Lines starting with ‘#MISSING:’ are
69 special comments documenting symbols that have disappeared.
70 Do not forget to check if old symbol versions need to be increased.
72 There is no way dpkg-gensymbols can warn about this. Blindly applying
73 the diff or assuming there is nothing to change if there is no diff,
74 without checking for such changes, can lead to packages with loose
75 dependencies that claim they can work with older packages they cannot
76 work with. This will introduce hard to find bugs with (partial)
77 upgrades.
78 Using #PACKAGE# substitution
80 In some rare cases, the name of the library varies between
81 architectures. To avoid hardcoding the name of the package in the
82 symbols file, you can use the marker #PACKAGE#. It will be replaced by
83 the real package name during installation of the symbols files.
84 Contrary to the #MINVER# marker, #PACKAGE# will never appear in a
85 symbols file inside a binary package.
86 Using symbol tags
88 Symbol tagging is useful for marking symbols that are special in some
89 way. Any symbol can have an arbitrary number of tags associated with
90 it. While all tags are parsed and stored, only some of them are
91 understood by dpkg-gensymbols and trigger special handling of the
92 symbols. See subsection Standard symbol tags for reference of these
93 tags.
94 Tag specification comes right before the symbol name (no whitespace is
96 allowed in between). It always starts with an opening bracket (, ends
97 with a closing bracket ) and must contain at least one tag. Multiple
98 tags are separated by the | character. Each tag can optionally have a
99 value which is separated form the tag name by the = character. Tag
100 names and values can be arbitrary strings except they cannot contain
101 any of the special ) | = characters. Symbol names following a tag
102 specification can optionally be quoted with either ' or " characters to
103 allow whitespaces in them. However, if there are no tags specified for
104 the symbol, quotes are treated as part of the symbol name which
105 continues up until the first space.
106 (tag1=i am marked|tag name with space)"tagged quoted symbol"@Base 1.0
108 (optional)tagged_unquoted_symbol@Base 1.0 1
109 untagged_symbol@Base 1.0
110 The first symbol in the example is named tagged quoted symbol and has
112 two tags: tag1 with value i am marked and tag name with space that has
113 no value. The second symbol named tagged_unquoted_symbol is only tagged
114 with the tag named optional. The last symbol is an example of the
115 normal untagged symbol.
116 Since symbol tags are an extension of the deb-symbols(5) format, they
118 can only be part of the symbols files used in source packages (those
119 files should then be seen as templates used to build the symbols files
120 that are embedded in binary packages). When dpkg-gensymbols is called
121 without the -t option, it will output symbols files compatible to the
122 deb-symbols(5) format: it fully processes symbols according to the
123 requirements of their standard tags and strips all tags from the
124 output. On the contrary, in template mode (-t) all symbols and their
125 tags (both standard and unknown ones) are kept in the output and are
126 written in their original form as they were loaded.
127 Standard symbol tags
129 optional
130 A symbol marked as optional can disappear from the library at
131 any time and that will never cause dpkg-gensymbols to fail.
132 However, disappeared optional symbols will continuously appear
133 as MISSING in the diff in each new package revision. This
134 behaviour serves as a reminder for the maintainer that such a
135 symbol needs to be removed from the symbol file or readded to
136 the library. When the optional symbol, which was previously
137 declared as MISSING, suddenly reappears in the next revision, it
138 will be upgraded back to the “existing” status with its minimum
139 version unchanged.
140 This tag is useful for symbols which are private where their
142 disappearance do not cause ABI breakage. For example, most of
143 C++ template instantiations fall into this category. Like any
144 other tag, this one may also have an arbitrary value: it could
145 be used to indicate why the symbol is considered optional.
146 arch=architecture-list
148 arch-bits=architecture-bits
149 arch-endian=architecture-endianness
150 These tags allow one to restrict the set of architectures where
151 the symbol is supposed to exist. The arch-bits and arch-endian
152 tags are supported since dpkg 1.18.0. When the symbols list is
153 updated with the symbols discovered in the library, all arch-
154 specific symbols which do not concern the current host
155 architecture are treated as if they did not exist. If an arch-
156 specific symbol matching the current host architecture does not
157 exist in the library, normal procedures for missing symbols
158 apply and it may cause dpkg-gensymbols to fail. On the other
159 hand, if the arch-specific symbol is found when it was not
160 supposed to exist (because the current host architecture is not
161 listed in the tag or does not match the endianness and bits), it
162 is made arch neutral (i.e. the arch, arch-bits and arch-endian
163 tags are dropped and the symbol will appear in the diff due to
164 this change), but it is not considered as new.
165 When operating in the default non-template mode, among arch-
167 specific symbols only those that match the current host
168 architecture are written to the symbols file. On the contrary,
169 all arch-specific symbols (including those from foreign arches)
170 are always written to the symbol file when operating in template
171 mode.
172 The format of architecture-list is the same as the one used in
174 the Build-Depends field of debian/control (except the enclosing
175 square brackets []). For example, the first symbol from the list
176 below will be considered only on alpha, any-amd64 and ia64
177 architectures, the second only on linux architectures, while the
178 third one anywhere except on armel.
179 (arch=alpha any-amd64 ia64)64bit_specific_symbol@Base 1.0
181 (arch=linux-any)linux_specific_symbol@Base 1.0
182 (arch=!armel)symbol_armel_does_not_have@Base 1.0
183 The architecture-bits is either 32 or 64.
185 (arch-bits=32)32bit_specific_symbol@Base 1.0
187 (arch-bits=64)64bit_specific_symbol@Base 1.0
188 The architecture-endianness is either little or big.
190 (arch-endian=little)little_endian_specific_symbol@Base 1.0
192 (arch-endian=big)big_endian_specific_symbol@Base 1.0
193 Multiple restrictions can be chained.
195 (arch-bits=32|arch-endian=little)32bit_le_symbol@Base 1.0
197 ignore-blacklist
199 dpkg-gensymbols has an internal blacklist of symbols that should
200 not appear in symbols files as they are usually only side-
201 effects of implementation details of the toolchain. If for some
202 reason, you really want one of those symbols to be included in
203 the symbols file, you should tag the symbol with
204 ignore-blacklist. It can be necessary for some low level
205 toolchain libraries like libgcc.
206 c++ Denotes c++ symbol pattern. See Using symbol patterns subsection
208 below.
209 symver Denotes symver (symbol version) symbol pattern. See Using symbol
211 patterns subsection below.
212 regex Denotes regex symbol pattern. See Using symbol patterns
214 subsection below.
215 Using symbol patterns
217 Unlike a standard symbol specification, a pattern may cover multiple
218 real symbols from the library. dpkg-gensymbols will attempt to match
219 each pattern against each real symbol that does not have a specific
220 symbol counterpart defined in the symbol file. Whenever the first
221 matching pattern is found, all its tags and properties will be used as
222 a basis specification of the symbol. If none of the patterns matches,
223 the symbol will be considered as new.
224 A pattern is considered lost if it does not match any symbol in the
226 library. By default this will trigger a dpkg-gensymbols failure under
227 -c1 or higher level. However, if the failure is undesired, the pattern
228 may be marked with the optional tag. Then if the pattern does not match
229 anything, it will only appear in the diff as MISSING. Moreover, like
230 any symbol, the pattern may be limited to the specific architectures
231 with the arch tag. Please refer to Standard symbol tags subsection
232 above for more information.
233 Patterns are an extension of the deb-symbols(5) format hence they are
235 only valid in symbol file templates. Pattern specification syntax is
236 not any different from the one of a specific symbol. However, symbol
237 name part of the specification serves as an expression to be matched
238 against name@version of the real symbol. In order to distinguish among
239 different pattern types, a pattern will typically be tagged with a
240 special tag.
241 At the moment, dpkg-gensymbols supports three basic pattern types:
243 c++
245 This pattern is denoted by the c++ tag. It matches only C++ symbols
246 by their demangled symbol name (as emitted by c++filt(1) utility).
247 This pattern is very handy for matching symbols which mangled names
248 might vary across different architectures while their demangled
249 names remain the same. One group of such symbols is non-virtual
250 thunks which have architecture specific offsets embedded in their
251 mangled names. A common instance of this case is a virtual
252 destructor which under diamond inheritance needs a non-virtual thunk
253 symbol. For example, even if _ZThn8_N3NSB6ClassDD1Ev@Base on 32bit
254 architectures will probably be _ZThn16_N3NSB6ClassDD1Ev@Base on
255 64bit ones, it can be matched with a single c++ pattern:
256 libdummy.so.1 libdummy1 #MINVER#
258 [...]
259 (c++)"non-virtual thunk to NSB::ClassD::~ClassD()@Base" 1.0
260 [...]
261 The demangled name above can be obtained by executing the following
263 command:
264 $ echo '_ZThn8_N3NSB6ClassDD1Ev@Base' | c++filt
266 Please note that while mangled name is unique in the library by
268 definition, this is not necessarily true for demangled names. A
269 couple of distinct real symbols may have the same demangled name.
270 For example, that's the case with non-virtual thunk symbols in
271 complex inheritance configurations or with most constructors and
272 destructors (since g++ typically generates two real symbols for
273 them). However, as these collisions happen on the ABI level, they
274 should not degrade quality of the symbol file.
275 symver
277 This pattern is denoted by the symver tag. Well maintained libraries
278 have versioned symbols where each version corresponds to the
279 upstream version where the symbol got added. If that's the case, you
280 can use a symver pattern to match any symbol associated to the
281 specific version. For example:
282 libc.so.6 libc6 #MINVER#
284 (symver)GLIBC_2.0 2.0
285 [...]
286 (symver)GLIBC_2.7 2.7
287 access@GLIBC_2.0 2.2
288 All symbols associated with versions GLIBC_2.0 and GLIBC_2.7 will
290 lead to minimal version of 2.0 and 2.7 respectively with the
291 exception of the symbol access@GLIBC_2.0. The latter will lead to a
292 minimal dependency on libc6 version 2.2 despite being in the scope
293 of the "(symver)GLIBC_2.0" pattern because specific symbols take
294 precedence over patterns.
295 Please note that while old style wildcard patterns (denoted by
297 "*@version" in the symbol name field) are still supported, they have
298 been deprecated by new style syntax "(symver|optional)version". For
299 example, "*@GLIBC_2.0 2.0" should be written as
300 "(symver|optional)GLIBC_2.0 2.0" if the same behaviour is needed.
301 regex
303 Regular expression patterns are denoted by the regex tag. They match
304 by the perl regular expression specified in the symbol name field. A
305 regular expression is matched as it is, therefore do not forget to
306 start it with the ^ character or it may match any part of the real
307 symbol name@version string. For example:
308 libdummy.so.1 libdummy1 #MINVER#
310 (regex)"^mystack_.*@Base$" 1.0
311 (regex|optional)"private" 1.0
312 Symbols like "mystack_new@Base", "mystack_push@Base",
314 "mystack_pop@Base" etc. will be matched by the first pattern while
315 e.g. "ng_mystack_new@Base" won't. The second pattern will match all
316 symbols having the string "private" in their names and matches will
317 inherit optional tag from the pattern.
318 Basic patterns listed above can be combined where it makes sense. In
320 that case, they are processed in the order in which the tags are
321 specified. For example, both
322 (c++|regex)"^NSA::ClassA::Private::privmethod\d\(int\)@Base" 1.0
324 (regex|c++)N3NSA6ClassA7Private11privmethod\dEi@Base 1.0
325 will match symbols "_ZN3NSA6ClassA7Private11privmethod1Ei@Base" and
327 "_ZN3NSA6ClassA7Private11privmethod2Ei@Base". When matching the first
328 pattern, the raw symbol is first demangled as C++ symbol, then the
329 demangled name is matched against the regular expression. On the other
330 hand, when matching the second pattern, regular expression is matched
331 against the raw symbol name, then the symbol is tested if it is C++ one
332 by attempting to demangle it. A failure of any basic pattern will
333 result in the failure of the whole pattern. Therefore, for example,
334 "__N3NSA6ClassA7Private11privmethod\dEi@Base" will not match either of
335 the patterns because it is not a valid C++ symbol.
336 In general, all patterns are divided into two groups: aliases (basic
338 c++ and symver) and generic patterns (regex, all combinations of
339 multiple basic patterns). Matching of basic alias-based patterns is
340 fast (O(1)) while generic patterns are O(N) (N - generic pattern count)
341 for each symbol. Therefore, it is recommended not to overuse generic
342 patterns.
343 When multiple patterns match the same real symbol, aliases (first c++,
345 then symver) are preferred over generic patterns. Generic patterns are
346 matched in the order they are found in the symbol file template until
347 the first success. Please note, however, that manual reordering of
348 template file entries is not recommended because dpkg-gensymbols
349 generates diffs based on the alphanumerical order of their names.
350 Using includes
352 When the set of exported symbols differ between architectures, it may
353 become inefficient to use a single symbol file. In those cases, an
354 include directive may prove to be useful in a couple of ways:
355 · You can factorize the common part in some external file and include
357 that file in your package.symbols.arch file by using an include
358 directive like this:
359 #include "packages.symbols.common"
361 · The include directive may also be tagged like any symbol:
363 (tag|...|tagN)#include "file-to-include"
365 As a result, all symbols included from file-to-include will be
367 considered to be tagged with tag ... tagN by default. You can use
368 this feature to create a common package.symbols file which includes
369 architecture specific symbol files:
370 common_symbol1@Base 1.0
372 (arch=amd64 ia64 alpha)#include "package.symbols.64bit"
373 (arch=!amd64 !ia64 !alpha)#include "package.symbols.32bit"
374 common_symbol2@Base 1.0
375 The symbols files are read line by line, and include directives are
377 processed as soon as they are encountered. This means that the content
378 of the included file can override any content that appeared before the
379 include directive and that any content after the directive can override
380 anything contained in the included file. Any symbol (or even another
381 #include directive) in the included file can specify additional tags or
382 override values of the inherited tags in its tag specification.
383 However, there is no way for the symbol to remove any of the inherited
384 tags.
385 An included file can repeat the header line containing the SONAME of
387 the library. In that case, it overrides any header line previously
388 read. However, in general it's best to avoid duplicating header lines.
389 One way to do it is the following:
390 #include "libsomething1.symbols.common"
392 arch_specific_symbol@Base 1.0
393 Good library management
395 A well-maintained library has the following features:
396 · its API is stable (public symbols are never dropped, only new
398 public symbols are added) and changes in incompatible ways only
399 when the SONAME changes;
400 · ideally, it uses symbol versioning to achieve ABI stability despite
402 internal changes and API extension;
403 · it doesn't export private symbols (such symbols can be tagged
405 optional as workaround).
406 While maintaining the symbols file, it's easy to notice appearance and
408 disappearance of symbols. But it's more difficult to catch incompatible
409 API and ABI change. Thus the maintainer should read thoroughly the
410 upstream changelog looking for cases where the rules of good library
411 management have been broken. If potential problems are discovered, the
412 upstream author should be notified as an upstream fix is always better
413 than a Debian specific work-around.
414
416 -Ppackage-build-dir
417 Scan package-build-dir instead of debian/tmp.
418 -ppackage
420 Define the package name. Required if more than one binary
421 package is listed in debian/control (or if there's no
422 debian/control file).
423 -vversion
425 Define the package version. Defaults to the version extracted
426 from debian/changelog. Required if called outside of a source
427 package tree.
428 -elibrary-file
430 Only analyze libraries explicitly listed instead of finding all
431 public libraries. You can use shell patterns used for pathname
432 expansions (see the File::Glob(3perl) manual page for details)
433 in library-file to match multiple libraries with a single
434 argument (otherwise you need multiple -e).
435 -Ifilename
437 Use filename as reference file to generate the symbols file that
438 is integrated in the package itself.
439 -O[filename]
441 Print the generated symbols file to standard output or to
442 filename if specified, rather than to debian/tmp/DEBIAN/symbols
443 (or package-build-dir/DEBIAN/symbols if -P was used). If
444 filename is pre-existing, its contents are used as basis for the
445 generated symbols file. You can use this feature to update a
446 symbols file so that it matches a newer upstream version of your
447 library.
448 -t Write the symbol file in template mode rather than the format
450 compatible with deb-symbols(5). The main difference is that in
451 the template mode symbol names and tags are written in their
452 original form contrary to the post-processed symbol names with
453 tags stripped in the compatibility mode. Moreover, some symbols
454 might be omitted when writing a standard deb-symbols(5) file
455 (according to the tag processing rules) while all symbols are
456 always written to the symbol file template.
457 -c[0-4]
459 Define the checks to do when comparing the generated symbols
460 file with the template file used as starting point. By default
461 the level is 1. Increasing levels do more checks and include all
462 checks of lower levels. Level 0 never fails. Level 1 fails if
463 some symbols have disappeared. Level 2 fails if some new symbols
464 have been introduced. Level 3 fails if some libraries have
465 disappeared. Level 4 fails if some libraries have been
466 introduced.
467 This value can be overridden by the environment variable
469 DPKG_GENSYMBOLS_CHECK_LEVEL.
470 -q Keep quiet and never generate a diff between generated symbols
472 file and the template file used as starting point or show any
473 warnings about new/lost libraries or new/lost symbols. This
474 option only disables informational output but not the checks
475 themselves (see -c option).
476 -aarch Assume arch as host architecture when processing symbol files.
478 Use this option to generate a symbol file or diff for any
479 architecture provided its binaries are already available.
480 -d Enable debug mode. Numerous messages are displayed to explain
482 what dpkg-gensymbols does.
483 -V Enable verbose mode. The generated symbols file contains
485 deprecated symbols as comments. Furthermore in template mode,
486 pattern symbols are followed by comments listing real symbols
487 that have matched the pattern.
488 -?, --help
490 Show the usage message and exit.
491 --version
493 Show the version and exit.
494
496 DPKG_GENSYMBOLS_CHECK_LEVEL
497 Overrides the command check level, even if the -c command-line
498 argument was given (note that this goes against the common
499 convention of command-line arguments having precedence over
500 environment variables).
501
503 https://people.redhat.com/drepper/symbol-versioning
504 https://people.redhat.com/drepper/goodpractice.pdf
505 https://people.redhat.com/drepper/dsohowto.pdf
506 deb-symbols(5), dpkg-shlibdeps(1).
5071.18.25 2018-06-26 dpkg-gensymbols(1)