Raise an error on syscall failure in tls_retry_write_records

[openssl.git] / doc / internal / man7 / build.info.pod

=pod

=head1 NAME

build.info - Building information files

=head1 SYNOPSIS

B<IF[>0|1B<]>

B<ELSIF[>0|1B<]>

B<ELSE>

B<ENDIF>

B<SUBDIRS=> I<dir> ...

B<PROGRAMS=> I<name> ...

B<LIBS=> I<name> ...

B<MODULES=> I<name> ...

B<SCRIPTS=> I<name> ...

B<DEPEND[>I<items>B<]=> I<otheritem> ...

B<GENERATE[>I<item>B<]=> I<generator> I<generator-args> ...

B<SOURCE[>I<item>B<]=> I<file> ...

B<SHARED_SOURCE[>I<item>B<]=> I<file> ...

B<DEFINE[>I<items>B<]=> I<name>[B<=>I<value>] ...

B<INCLUDE[>I<items>B<]=> I<dir> ...

B<$>I<VARIABLE>B<=>I<value>

=head1 DESCRIPTION

OpenSSL's build system revolves around three questions:

=over 4

=item What to build for?

This is about choice of platform (combination of hardware, operating
system, and toolchain).

=item What to build?

This is about having all the information on what needs to be built and
from what.

=item How to build it?

This is about build file generation.

=back

This document is all about the second item, "What to build?", and most
of all, how to specify that information.

For some terms used in this document, please see the L</GLOSSARY> at
the end.

=head2 F<build.info> files

F<build.info> files are meta data files for OpenSSL's built file
generators, and are used to specify exactly what end product files
(programs, libraries, modules or scripts) are to be produced, and from
what sources.

Intermediate files, such as object files, are seldom referred to at
all.  They sometimes can be, if there's a need, but this should happen
very rarely, and support for that sort of thing is added on as-needed
basis.

Any time a directory or file is expected in a statement value, Unix
syntax must be used, which means that the slash C</> must be used as
the directory separator.

=head2 General syntax

=head3 Comments

Comments are any line that start with a hash sign (C<#>).  The hash
sign may be preceded by any number of horizontal spaces.

=head3 Filenames

F<build.info> files are platform agnostic.  This means that there is
some information in them that is representative rather than specific.

This is particularly visible with end product names, they work more
like a tag than as the actual filename that's going to be produced.
This is because different platforms have different decorations on
different types of files.

For example, if we say that we want to produce a program C<foo>, it
would look like this:

    PROGRAM=foo

However, the program filename may end up being just C<foo> (typical
for Unix), or C<foo.exe> (typical for Windows), or even C<BLAH$FOO.EXE>
(possible on VMS, depending on policy).

These platform specific decorations are not the concern of
F<build.info> files.  The build file generators are responsible for
transforming these platform agnostic names to their platform specific
counterparts.

=head3 Statements

With the exception of variables and conditions, the general statement
syntax is one of:

=over 4

=item B<I<KEYWORD>> B<=> I<value> ...

=item B<I<KEYWORD>[>I<items>B<]> B<=> I<value> ...

=back

Every B<I<KEYWORD>> represents some particular type of information.

The first form (sometimes called "plain statement") is used to specify
information on what end products need to be built, for example:

    PROGRAMS=foo bar
    LIBS=libpoly libcookie
    MODULES=awesome-plugin
    SCRIPTS=tool1 tool2
    SUBDIRS=dir1 dir2

This says that we want to build programs C<foo> and C<bar>, the
libraries C<libpoly> and C<libcookie>, an awesome plugin module
C<awesome-plugin>, a couple of scripts C<tool1> and C<tool2>, and
finally that there are more F<build.info> files in subdirectories
C<dir1> and C<dir2>.

The second form (sometimes called "indexed statement") is used to
specify further details for existing items, for example:

    SOURCE[foo]=foo.c details.c
    DEPEND[foo]=libcookie

This says that the program C<foo> is built from the source files
F<foo.c> and F<details.c>, and that it depends on the library
C<libcookie> (in other words, the library will be included when
linking that program together).

Multiple space separated items are allowed too:

    SOURCE[foo]=foo.c
    SOURCE[details]=details.c
    DEPEND[foo details]=libcookie

For any indexed statement for which the items haven't been specified
through any plain statement, or where the items exists but the indexed
statement does not apply, the value is simply ignored by the build
file generators.

=head3 Statement attributes

Some statements can have attributes added to them, to allow for
variations on how they are treated.

=over 4

=item B<I<KEYWORD>{> I<attrib> | I<attrib>B<=>I<attrib-value> [,...]B<}>
B<=> I<value> ...

=item B<I<KEYWORD>[>I<items>B<]{> I<attrib> | I<attrib>B<=>I<attrib-value>
[,...]B<}> B<=> I<value> ...

=back

Attributes are passed as they are to the build file generators, and
the exact interpretation of those attributes is entirely up to them
(see L</Known attributes> below for details).

A current example:

    LIBS{noinst,has_main}=libtestutil.a

This says that the static library C<libtestutil.a> should not be
installed (C<noinst>), and that it includes an object file that has
the C<main> symbol (C<has_main>).  Most platforms don't need to know
the latter, but there are some where the program linker will not look
for C<main> in libraries unless it's explicitly told so, so this is
way to tell the build file generator to emit the necessary command
options to make that happen.

Attributes are accumulated globally.  This means that a library could
be given like this in different places:

    # Location 1
    LIBS=libwhatever

    # Location 2
    LIBS{noinst}=libwhatever

    # Location 3
    LIBS{has_main}=libwhatever

The end result is that the library C<libwhatever> will have the
attributes C<noinst> and C<has_main> attached to it.

=head3 Quoting and tokens

Statement values are normally split into a list of tokens, separated
by spaces.

To avoid having a value split up into several tokens, they may be
quoted with double (C<">) or single (C<'>) quotes.

For example:

    PROGRAMS=foo "space cadet" bar

This says that we sant to build three programs, C<foo>, C<space cadet>
and C<bar>.

=head3 Conditionals

F<build.info> files include a very simple condition system, involving
the following keywords:

=over 4

=item B<IF[>0|1B<]>

=item B<ELSIF[>0|1B<]>

=item B<ELSE>

=item B<ENDIF>

=back

This works like any condition system with similar syntax, and the
condition value in B<IF> and B<ELSIF> can really be any literal value
that perl can interpret as true or false.

Conditional statements are nesting.

In itself, this is not very powerful, but together with L</Perl nuggets>,
it can be.

=head3 Variables

F<build.info> handles simple variables.  They are defined by
assignment:

=over 4

=item B<$>I<NAME> B<=> I<value>

=back

These variables can then be used as part of any statement value or
indexed statement item.  This should be used with some care, as
I<variables are expanded into their values before the value they are
part of is tokenized>.

I<Variable assignment values are not tokenized.>

Variable references can be one of:

=over 4

=item B<$>I<NAME> or B<${>I<NAME>B<}>

Simple reference; the variable reference is replaced with its value,
verbatim.

=item B<${>I<NAME>B</>I<str>B</>I<subst>B<}>

Substitution reference; the variable reference is replaced with its
value, modified by replacing all occurrences of I<str> with I<subst>.

=back

=head2 Scope

Most of the statement values are accumulated globally from all the
F<build.info> files that are digested.  There are two exceptions,
F<build.info> variables and B<SUBDIRS> statement, for which the scope
is the F<build.info> file they are in.

=head2 Perl nuggets

Whenever a F<build.info> file is read, it is passed through the Perl
template processor L<OpenSSL::Template>, which is a small extension of
L<Text::Template>.

Perl nuggets are anything between C<{-> and C<-}>, and whatever the
result from such a nugget is, that value will replace the nugget in
text form.  This is useful to get dynamically generated F<build.info>
statements, and is most often seen used together with the B<IF> and
B<ELSIF> conditional statements.

For example:

    IF[{- $disabled{something} -}]
      # do whatever's needed when "something" is disabled
    ELSIF[{- $somethingelse eq 'blah' -}]
      # do whatever's needed to satisfy this condition
    ELSE
      # fallback
    ENDIF

Normal Perl scope applies, so it's possible to have an initial perl
nugget that sets diverse global variables that are used in later
nuggets.  Each nugget is a Perl block of its own, so B<my> definitions
are only in scope within the same nugget, while B<our> definitions are
in scope within the whole F<build.info> file.

=head1 REFERENCE

=head2 Conditionals

=over 4

=item B<IF[>0|1B<]>

If the condition is true (represented as C<1> here), everything
between this B<IF> and the next corresponding B<ELSIF> or B<ELSE>
applies, and the rest until the corresponding B<ENDIF> is skipped
over.

If the condition is false (represented as C<0> here), everything
from this B<IF> is skipped over until the next corresponding B<ELSIF>
or B<ELSE>, at which point processing continues.

=item B<ELSE>

If F<build.info> statements have been skipped over to this point since
the corresponding B<IF> or B<ELSIF>, F<build.info> processing starts
again following this line.

=item B<ELSIF[>0|1B<]>

This is B<ELSE> and B<IF> combined.

=item B<ENDIF>

Marks the end of a conditional.

=back

=head2 Plain statements

=over 4

=item B<SUBDIRS=> I<dir> ...

This instructs the F<build.info> reader to also read the F<build.info>
file in every specified directory.  All directories should be given
relative to the location of the current F<build.info> file.

=item B<PROGRAMS=> I<name> ...

Collects names of programs that should be built.

B<PROGRAMS> statements may have attributes, which apply to all the
programs given in such a statement.  For example:

    PROGRAMS=foo
    PROGRAMS{noinst}=bar

With those two lines, the program C<foo> will not have the attribute
C<noinst>, while the program C<bar> will.

=item B<LIBS=> I<name> ...

Collects names of libraries that should be built.

The normal case is that libraries are built in both static and shared
form.  However, if a name ends with C<.a>, only the static form will
be produced.

Similarly, libraries may be referred in indexed statements as just the
plain name, or the name including the ending C<.a>.  If given without
the ending C<.a>, any form available will be used, but if given with
the ending C<.a>, the static library form is used unconditionally.

B<LIBS> statements may have attributes, which apply to all the
libraries given in such a statement.  For example:

    LIBS=libfoo
    LIBS{noinst}=libbar

With those two lines, the library C<libfoo> will not have the
attribute C<noinst>, while the library C<libbar> will.

=item B<MODULES=> I<name>

Collects names of dynamically loadable modules that should be built.

B<MODULES> statements may have attributes, which apply to all the
modules given in such a statement.  For example:

    MODULES=foo
    MODULES{noinst}=bar

With those two lines, the module C<foo> will not have the attribute
C<noinst>, while the module C<bar> will.

=item B<SCRIPTS=> I<name>

Collects names of scripts that should be built, or that just exist.
That is how they differ from programs, as programs are always expected
to be compiled from multiple sources.

B<SCRIPTS> statements may have attributes, which apply to all the
scripts given in such a statement.  For example:

    SCRIPTS=foo
    SCRIPTS{noinst}=bar

With those two lines, the script C<foo> will not have the attribute
C<noinst>, while the script C<bar> will.

=back

=head2 Indexed statements

=over 4

=item B<DEPEND[>I<items>B<]> B<=> I<file> ...

Collects dependencies, where I<items> depend on the given I<file>s.

As a special case, the I<items> may be empty, for which the build file
generators should make the whole build depend on the given I<file>s,
rather than the specific I<items>.

The I<items> may be any program, library, module, script, or any
filename used as a value anywhere.

The I<items> may also be literal build file targets.  Those are
recognised by being surrounded be vertical bars (also known as the
"pipe" character), C<|>.  For example:

    DEPEND[|tests|]=fipsmodule.cnf

B<DEPEND> statements may have attributes, which apply to each
individual dependency in such a statement.  For example:

    DEPEND[libfoo.a]=libmandatory.a
    DEPEND[libfoo.a]{weak}=libbar.a libcookie.a

With those statements, the dependency between C<libfoo.a> and
C<libmandatory.a> is strong, while the dependency between C<libfoo.a>
and C<libbar.a> and C<libcookie.a> is weak.  See the description of
B<weak> in L</Known attributes> for more information.

B<DEPEND> is a bit more involving when used with I<item>s that are
generated with B<GENERATE>.  This is described more in depth below.

=item B<GENERATE[>I<item>B<]> B<=> I<generator> I<generator-arg> ...

This specifies that the I<item> is generated using the I<generator>
with the I<generator-arg>s as arguments, plus the name of the output
file as last argument.

The build file generators must be able to recognise the I<generator>.
Currently, they at least recognise files ending in C<.pl>, and will
execute them to generate the I<item>, and files ending in C<.in>,
which will be used as input for L<OpenSSL::Template> to generate
I<item> (in other words, we use the exact same style of
L</Perl nuggets> mechanism that is used to read F<build.info> files).

For I<generator>s where this is applicable, any B<INCLUDE> statement
for the same I<item> will be given to the I<generator> as its
inclusion directories.

Likewise, For I<generator>s where this is applicable, any B<DEPEND>
statement for the same I<item> will be given to the I<generator> as an
extra file or module to load, where this is applicable.

=over 4

=item The B<DEPEND> statement may be problematic:

Depending on what generator is used, a B<DEPEND> statement also acts
as an B<INCLUDE> statement for the directory where the I<file> is
located.  In some cases, that's not quite feasible, because a module
isn't meant to be loaded by filename only and may require a nondefault
separation between the implied inclusion directory and the intended module
name.

=item ... but there is a solution:

To enable that sort of separation, B<DEPEND> can use a slightly
different I<file> syntax, that looks like this:

B<DEPEND[>I<items>B<]> B<=> I<dir>|I<module>

The I<module> must be specified in a way that makes sense for the generator.
For example, when the generator implies perl (ends with C<.in>) and depends
on the module F<OpenSSL::foo> - a.k.a. F<OpenSSL/foo.pm> - which lives in
F<util/perl>, it feasible to have something like this:

    GENERATE[something.c]=something.c.in
    DEPEND[something.c]=util/perl|OpenSSL/foo.pm

=back

=item B<SOURCE[>I<item>B<]> B<=> I<file> ...

Collects filenames that will be used as source files for I<item>.

The I<item> must be a singular item, and may be any program, library,
module or script given with B<PROGRAMS>, B<LIBS>, B<MODULES> and
B<SCRIPTS>.

Static libraries may be sources.  In that case, its object files are
used directly when building I<item> instead of relying on library
dependency and symbol resolution (through B<DEPEND> statements).

B<SOURCE> statements may have attributes, which apply to each
individual dependency in such a statement.  For example:

    SOURCE[prog]=prog_a.c
    SOURCE[prog]{check}=prog_b.c prog_c.c

With those statements, the association between C<prog> and C<prog_a.c>
comes with no extra attributes, while the association between C<prog>
and C<prog_b.c> as well as C<prog_c.c> comes with the extra attribute
C<check>.

=item B<SHARED_SOURCE[>I<item>B<]> B<=> I<file> ...

Collects filenames that will be used as source files for I<item>.

The I<item> must be a singular item, and may be any library or module
given with B<LIBS> or B<MODULES>.  For libraries, the given filenames
are only used for their shared form, so if the item is a library name
ending with C<.a>, the filenames will be ignored.

B<SHARED_SOURCE> statements may have attributes, just as B<SOURCE>
statements.

=item B<DEFINE[>I<items>B<]> B<=> I<name>[B<=>I<value>] ...

Collects I<name> / I<value> pairs (or just I<name> with no defined
value if no I<value> is given) associated with I<items>.

The build file generators will decide what to do with them.  For
example, these pairs should become C macro definitions whenever a
C<.c> file is built into an object file.

=item B<INCLUDE[>I<items>B<]> B<=> I<dir> ...

Collects inclusion directories that will be used when building the
I<items> components (object files and whatever else).  This is used at
the discretion of the build file generators.

=back

=head2 Known attributes

Note: this will never be a complete list of attributes.

=over 4

=item B<noinst>

This is used to specify that the end products this is set for should
not be installed, that they are only internal.  This is applicable on
internal static libraries, or on test programs.

=item B<misc>

This is used with B<SCRIPTS>, to specify that some scripts should be
installed in the "misc" directory rather than the normal program
directory.

=item B<engine>

This is used with B<MODULES>, to specify what modules are engines and
should be installed in the engines directory instead of the modules
directory.

=item B<weak>

This is used with B<DEPEND> where libraries are involved, to specify
that the dependency between two libraries is weak and is only there to
infer order.

Without this attribute, a dependency between two libraries, expressed
like this, means that if C<libfoo.a> appears in a linking command
line, so will C<libmandatory.a>:

    DEPEND[libfoo.a]=libmandatory.a

With this attribute, a dependency between two libraries, expressed
like this, means that if I<both> C<libfoo.a> and C<libmandatory.a>
appear in a linking command line (because of recursive dependencies
through other libraries), they will be ordered in such a way that this
dependency is maintained:

    DEPEND[libfoo.a]{weak}=libfoo.a libcookie.a

This is useful in complex dependency trees where two libraries can be
used as alternatives for each other.  In this example, C<lib1.a> and
C<lib2.a> have alternative implementations of the same thing, and
C<libmandatory.a> has unresolved references to that same thing, and is
therefore depending on either of them, but not both at the same time:

    DEPEND[program1]=libmandatory.a lib1.a
    DEPEND[program2]=libmandatory.a lib2.a
    DEPEND[libmandatory]{weak}=lib1.a lib2.a

=back

=head1 GLOSSARY

=over 4

=item "build file"

This is any platform specific file that describes the complete build,
with platform specific commands.  On Unix, this is typically
F<Makefile>; on VMS, this is typically F<descrip.mms>.

=item "build file generator"

Perl code that generates build files, given configuration data and
data collected from F<build.info> files.

=item "plain statement"

Any F<build.info> statement of the form B<I<KEYWORD>>=I<values>, with
the exception of conditional statements and variable assignments.

=item "indexed statement"

Any F<build.info> statement of the form B<I<KEYWORD>[>I<items>B<]=>I<values>,
with the exception of conditional statements.

=item "intermediate file"

Any file that's an intermediate between a source file and an end
product.

=item "end product"

Any file that is mentioned in the B<PROGRAMS>, B<LIBS>, B<MODULES> or
B<SCRIPTS>.

=back

=head1 SEE ALSO

For OpenSSL::Template documentation,
C<perldoc -o man util/perl/OpenSSL/Template.pm>

L<Text::Template|https://metacpan.org/pod/Text::Template>

=head1 COPYRIGHT

Copyright 2019-2023 The OpenSSL Project Authors. All Rights Reserved.

Licensed under the Apache License 2.0 (the "License").  You may not use this
file except in compliance with the License.  You can obtain a copy in the file
LICENSE in the source distribution or at
L<https://www.openssl.org/source/license.html>.

=cut