Annotation of wikisrc/pkgsrc/gcc.mdwn, revision 1.9

1.1       gdt         1: On many systems pkgsrc supports, gcc is the standard compiler.  In
                      2: general, different versions of each OS have different gcc versions,
1.2       gdt         3: and some packages require newer gcc versions, in order to support
1.1       gdt         4: newer language standards (e.g. c++11, written in the style of
                      5: USE_LANGUAGES), or because older versions don't work (infrequently).
                      7: This page discusses issues related to version selection, and intends
                      8: to be a design document for how pkgsrc should address this problem, to
                      9: be converted into historical design rationale once implemented.  It
                     10: freely takes content from extensive mailinglist discussions, and
                     11: attempts to follow the rough consensus that has emerged.
                     13: ## Base system gcc vs pkgsrc gcc
                     15: Systems using gcc (e.g. NetBSD) have a compiler as /usr/bin/gcc, and
                     16: this is usable by pkgsrc without any bootstrapping activity.  One can
                     17: build gcc versions (typically newer versions) from pkgsrc, resulting
                     18: in a compiler within ${PREFIX}, e.g. /usr/pkg/gcc6/bin/gcc.  This
                     19: compiler can then be used to compile other packages.
1.9     ! gdt        21: The issue with using base system gcc is typically that it is too old,
1.6       gdt        22: such as gcc 4.5 with NetBSD 6, which cannot compile c++11.  Another
                     23: example is gcc 4.8 with NetBSD 7.  While this can compile most c++11
                     24: programs, it cannot be used for firefox or glibmm (and therefore any
                     25: package that links against glibmm).
1.1       gdt        26: 
                     27: Issues when using pkgsrc gcc are that
1.9     ! gdt        29:   - on some platforms, pkgsrc gcc does not build and work
1.1       gdt        30:   - it must be bootstrapped, requiring compiling a number of packages
                     31:     with the system compiler
                     32:   - C++ packages that are linked together should be built with the
                     33:     same compiler, because the standard library ABI is not necessarily
                     34:     the same for each compiler version
                     35:   - While C packages can be built with mixed versions, the binary
                     36:     should be linked with the higher version because the support
                     37:     library is backwards compatible but not forward compatible.
                     39: ## Specific constraints and requirements
                     41: This section attempts to gather all the requirements.
                     43:   - By default, pkgsrc should be able to build working packages, even
                     44:     for packages that need a newer compiler than that provided in the
                     45:     base system.
                     47:   - The set of packages that are needed when building a bootstrap
                     48:     compiler should be minimized.
                     50:   - All packages that use C should have final linking with the highest
                     51:     version used in any included library.
1.2       gdt        53:   - All packages that use C++ should be built with the same compiler
1.3       gdt        54:     version.  Because these in the general case may include C, the
                     55:     version used for C++ must be at least as new as the version used
                     56:     for any used C package.
1.2       gdt        57: 
1.1       gdt        58:   - pkgsrc should avoid building gcc unless it is more or less
                     59:     necessary to build packges.  (As an example, if the base system
                     60:     gcc can build c99 but not c++11, building a c99-only program
                     61:     should not trigger building a gcc version adequate for c++11.)
1.2       gdt        63:   - The compiler selection logic should work on NetBSD 6 and newer,
                     64:     and other systems currently supported by pkgsrc, including in-use
                     65:     LTS GNU/Linux systems.  It should work on systems that default to
                     66:     clang, when set to use GCC, at least as well as the current
                     67:     scheme.  It is desirable for this logic to work on NetBSD 5.
1.1       gdt        68: 
1.9     ! gdt        69:   - All systems should work at least as well as they do before
        !            70:     implementation of new compiler selection logic.
        !            71: 
1.1       gdt        72:   - The compiler selection logic should be understandable and not brittle.
                     74: ## Design
                     76: The above requirements could in theory be satisfied in many ways, but
1.2       gdt        77: most of them are too complicated.  We present a design that aims to be
                     78: sound while mimimizing complexity.
1.1       gdt        79: 
                     80:   - Packages declare what languages they need, with c++, c++11, and
1.2       gdt        81:     c++14 being expressed differently.  (This is exactly current
                     82:     practice and just noted for completeness.)
1.1       gdt        83: 
                     84:   - The package-settable variable GCC_REQD will be used only when a
                     85:     compiler that generally can compile the declared language version
1.2       gdt        86:     is insufficient.  These cases are expected to be relatively rare;
                     87:     an example is firefox that is in c++ (but not c+11) and needs gcc
                     88:     4.9.
1.1       gdt        89: 
                     90:   - A user-settable variable PKGSRC_GCC_VERSION will declare the
1.9     ! gdt        91:     version of gcc to be used for C programs, with an OS-,
        !            92:     version- and architeture- specific default.
1.1       gdt        93: 
1.6       gdt        94:   - A user-settable variable PKGSRC_GXX_VERSION will declare the
1.9     ! gdt        95:     version of gcc to be used for all C++ programs, again with an OS-,
        !            96:     version- and architeture-specific default.  It must be at least
1.6       gdt        97:     PKGSRC_GCC_VERSION.
1.1       gdt        98: 
1.9     ! gdt        99:   - If PKGSRC_GCC_VERSION and PKGSRC_GXX_VERSION are not set, the
        !           100:     system will behave much as before.  As a possible exception,
        !           101:     builds may still fail if the required version is greater than the
        !           102:     base system version.  So far the only known reason to avoid
        !           103:     setting these variable is if pkgsrc gcc cannot be built.
        !           104: 
1.1       gdt       105:   - Each of c99, c++, c++11, and c++14 will be associated with a
                    106:     minimum gcc version, such that almost all programs declaring that
                    107:     language can be built with that version.  (This avoids issues of
                    108:     strict compliance with c++11, which requires a far higher version
                    109:     of gcc than the version required to compile almost all actual
                    110:     programs in c++11.)
                    112:   - The minimum version inferred from the language tag will be
                    113:     combined with any GCC_REQD declarations to find a minimum version
                    114:     for a specific package.  If that is greater than
                    115:     PKGSRC_GCC_VERSION (programs using only C) or PKGSRC_GXX_VERSION,
                    116:     package building will fail.  We call the resulting
                    117:     PKGSRC_GCC_VERSION or PKGSRC_GXX_VERSION the chosen version.
1.3       gdt       119:   - When building a program using C or C++, if the chosen version is
                    120:     not provided by the base system, and the chosen version is not
1.1       gdt       121:     installed via pkgsrc, then it (and its dependencies) will be built
                    122:     from pkgsrc in a special bootstrap mode.  When building in
                    123:     bootstrap mode, the version selection logic is ignored and the
                    124:     base system compiler is used.  Consistency and reproducible builds
                    125:     require that a package built with the normal prefix must be the
                    126:     same whether built because of compiler bootstrapping or normal
                    127:     use.
                    129:     There are thus two choices for dealing with bootstrapping.  One is
                    130:     to use a distinct prefix, which will ensure that all packages that
                    131:     are part of the compiler bootstrap will not be linked into normal
                    132:     pkgsrc programs.  This implies that any dependencies of gcc may
                    133:     exist twice, once in bootstrap mode and once if built normally.  A
                    134:     gcc version itself will be built twice, if it is desired for
                    135:     regular use.  This double building and the complexity of a second
                    136:     prefix are the negatives of this approach.
                    138:     The other choice is to mark gcc and all depending packages as used
                    139:     for compiler bootstrapping, and to always build those with the
                    140:     base compiler.  We use the package-settable variable
                    141:     PKGSRC_GCC_BOOTSTRAP=yes to denote this.  The negative with this
                    142:     approach is possible inconsistency with gcc's dependencies being
                    143:     built with the base compiler and used later.
1.9     ! gdt       145:     As an alternative, we store lists of bootstrap packages in a
        !           146:     variable, because it will vary with OS and version, and with
        !           147:     PREFER_PKGSRC settings.
        !           148: 
        !           149:     As a third alternative, we pass a GCC_BOOTSTRAPPING variable
        !           150:     recursively.  This is easier but less consistent.
        !           151: 
1.2       gdt       152:   - We hope that the chosen version can be built using the base system
                    153:     version, and hope to avoid multi-stage bootstrapping.
1.1       gdt       155:   - We expect that any program containing C++ will undergo final
                    156:     linking with a C++ compiler.  This is not a change from the
                    157:     current situation.
                    159: ## Remaining issues
1.6       gdt       161: ### gcc dependencies introduction
1.1       gdt       162: 
                    163: Because gcc can have dependencies, there could be packages built with
                    164: the system compiler that are then later used with the chosen version.
                    165: For now, we defer worrying about these problems (judging that they
                    166: will be less serious than the current situation where all c++11
                    167: programs fail to build on NetBSD 6).
1.9     ! gdt       169: \todo: Perhaps change gcc 4.8 and 4.9 to enable gcc-inplace-math by
        !           170: default.  Perhaps decide that if we want to build gcc, we want to
        !           171: build 5 or 6, and 4.9 is no longer of interest as a bootstrap target.
1.6       gdt       172: 
1.1       gdt       173: \todo: Analyze what build-time and install-time dependencies actually
1.6       gdt       174: exist.  Include old GNU/Linux in this analysis.
                    176: \todo: Consider if dropping nls would help.  (On NetBSD, it seems that
                    177: base system libraries are used, so it would not help.)
1.7       gdt       179: \todo: Consider failing if optins that we want one way are another,
                    180: when bootstrapping.
1.6       gdt       182: ### managing gcc dependencies
                    184: There are multiple paths forward.
                    186: \todo Choose one.  Straw proposal is "Don't worry" and recursive
                    187: variable for the initial implementation.
                    189: #### Separate prefix
                    191: Build compilers in a separate prefix, or a subprefix, so that the
                    192: compiler and the packages needed to build it will not be used by any
                    193: normal packages.  This completely avoids the issue of building a
                    194: package one way in bootstrap and another not in bootstrap, at the cost
                    195: of two builds and writing the separate-prefix code.
                    197: #### Don't worry
                    199: Don't worry that packages used to bootstrap the needed compiler are
                    200: compiled with an older compiler.  Don't worry that they might be
                    201: different depending on build order.  If we have an actual problem,
                    202: deal with it.  This requires choosing an approach to omit compiler
                    203: selection logic when building the compiler:
                    205: ##### Mark bootstrap packages
                    207: Mark packages used to build gcc as PKGSRC_GCC_BOOTSTRAP=yes.
                    208: Conditionalize this on OPSYS if necessary.  Don't force the compiler
                    209: if this is set.
1.1       gdt       210: 
1.8       gdt       211: Alternatively, manage a per-OS list of packages in a central mk file.
1.6       gdt       213: ##### Pass a recursive variable
                    215: As above, but set PKGSRC_GCC_BOOTSTRAP=yes in the evniroment of the
                    216: call to build the compiler, so that all dependencies inherit
                    217: permission to skip compiler selection logic.  (Alternatively, use some
                    218: other mechanism such as passing a make variable explicitly.)
1.1       gdt       219: 
1.3       gdt       220: ### Differing GCC and GXX versions
                    222: Perhaps it is a mistake to allow the chosen GCC and GXX versions to
                    223: differ.  If we require them to be the same, then essentially all
                    224: systems with a base system compiler older than gcc 5 will have to
                    225: bootstrap the compiler.  For now, we allow them to differ and will
                    226: permit the defaults to differ.
1.6       gdt       228: ### gcc versions and number of buildable packages
                    230: A gcc version that is too old will not build a number of packages.
                    231: Anything older than 4.8 fails for c++11.  4.8 fails on some c++11
                    232: packages, such as firefox and glibmm.
                    234: A version that is too new also fails to build packages.  Analyses
                    235: posted to tech-pkg indicate that 5 is close to 4.9 in the number of
                    236: packages built, and that moving to 6 causes hundreds of additional
                    237: failures.
                    239: Therefore, the current answer to "What is the best version to use" is
                    240: 5.
1.1       gdt       242: ### Default versions for various systems
1.3       gdt       244: Note that if for any particular system's set of installed packages (or
                    245: bulk build), a newer gcc has to be built, it does not hurt to have
                    246: built it earlier.
1.1       gdt       247: 
                    248: When the base system is old (e.g., gcc 4.5 in NetBSD 6, or 4.1, in
                    249: NetBSD 5), then it is clear that a newer version must be built.  For
                    250: these, PKGSRC_GXX_VERSION should default to a newish gcc, avoiding
1.4       gdt       251: being so new as to cause building issues.  Currently, gcc5 is probably
                    252: a good choice, with gcc6 compiling significantly but not vastly fewer
                    253: packages.  PKGSRC_GCC_VERSION should probably default to the system
                    254: version if it can build all C99 programs, or match PKGSRC_GXX_VERSION,
                    255: if the system version is too old.  Perhaps gcc 4.5 would be used, but
                    256: 4.1 not used.  \todo Discuss.
1.1       gdt       257: 
                    258: When the base system is almost new enough, the decision about the
                    259: default is more complicated.  A key example is gcc 4.8, found in
1.2       gdt       260: NetBSD 7.  Firefox requires gcc 4.9, and all programs using c++14 also
1.1       gdt       261: need a newer version.  One options is to choose 4.8, resulting in
                    262: firefox failing, as well as all c++14 programs.  Another is to choose
                    263: 4.9, but this makes little sense because c++14 programs will still
                    264: fail, and the general rule of moving to the most recent
                    265: generally-acceptable version applies, which currently leads to gcc6.
                    266: This is in effect a declaration that "almost new enough" does not
                    267: count as new enough.  Thus the plan for NetBSD 7 is to set
1.4       gdt       268: PKGSRC_GCC_VERSION to 4.8 and PKGSRC_GXX_VERSION to 5.
1.1       gdt       269: 
1.4       gdt       270: When the base system is new enough, e.g. gcc 5, 6 or 7 it should
                    271: simply be used.  By "new enough", we mean that almost no programs in
1.6       gdt       272: pkgsrc fail to build with it (because it is too old), which implies
                    273: that it supports (almost all) C++14 programs.  Our current definiton
                    274: of new enough is gcc 5.
1.1       gdt       275: 
1.2       gdt       276: ### Limited mixed versions
                    278: One approach would be to allow limited mixed versions, where
                    279: individual programs could force a specific version to be bootstrapped
                    280: and used, so that e.g. firefox could use 4.9 even though most programs
                    281: use 4.8, which is what happens now on NetBSD 7.  This would rely on
                    282: being able to link c++ with 4.9 including some things built with 4.8
                    283: (which is done presently).  However, this approach would become
                    284: unsound with a library rather than an end program.  We reject this as
                    285: too much complexity for avoiding building a newer compiler in limited
                    286: situations.
1.1       gdt       288: ### Fortran
                    290: Fortran support is currently somewhat troubled..  It seems obvious to
                    291: extend to PGKSRC_GFORTRAN_VERSION, and have that match
                    292: PKGSRC_GCC_VERSION or PKGSRC_GXX_VERSION, but the Fortran situation is
1.6       gdt       293: not worsened by the above design.
                    295: When building a gcc version, we get gfortran.  Perhaps, because of
                    296: fortran, we should require a single version, vs a C and a C++ version.
                    298: \todo Discuss.
1.9     ! gdt       300: ### C++ libraries used by C programs
1.6       gdt       301: 
                    302: The choice of one version for C++ and one for C (e.g. 5, 4.8 on
                    303: netbsd-7) breaks down if a C program links against a library that is
                    304: written in C++ but provides a C API, because we still need the C++
                    305: version's stdlib.
                    307: \todo Define a variable for such packages to have in their buildlink3,
                    308: which will not add c++ to USE_LANGUAGES but will force
                    309: PKGSRC_GXX_VERSION to be used.  Or decide that this is a good reason
                    310: to really just have one compiler version.
1.1       gdt       311: 
                    312: ## Path forward
1.9     ! gdt       314: (This assumes per-package marking of bootstrap packages, but is
        !           315: reasonably obviously extended to the other schemes.)
        !           316: 
1.1       gdt       317:  - Modify all gcc packages to have minimal dependencies, and to add
                    318:    PKGSRC_GCC_BOOTSTRAP.
                    320:  - Modify the compiler selection logic to do nothing if
                    321:    PKGSRC_GCC_BOOTSTRAP is set.
                    323:  - Modify the compiler selection logic for LANGUAGES= to fail if
                    324:    PKGSRC_GCC_VERSION/PKGSRC_GXX_VERSION is not new enough.
1.9     ! gdt       326:  - Modify the compiler selection logic for GCC_REQD to fail if
        !           327:    PKGSRC_GCC_VERSION/PKGSRC_GXX_VERSION is not new enough.
1.1       gdt       328: 
                    329:  - Decide on defaults.  The straw proposal is that PKGSRC_GCC_VERSION
1.5       gdt       330:    is the base system version if >= 4.5 (or 4.4?), and otherwise 5,
1.1       gdt       331:    and that PKGSRC_GXX_VERSION is the base system version if >= 5, and
1.9     ! gdt       332:    otherwise 5.  Implement these in as they are tested.
1.3       gdt       333: 
                    334: ### Later steps
1.9     ! gdt       336:  - Address fortran.  Probably add PKGSRC_GFORTRAN_VERSION, after
        !           337:    determining how Fortran, C and C++ interact with library ABI
        !           338:    compatibility.

CVSweb for NetBSD wikisrc <> software: FreeBSD-CVSweb