Skip to content

Latest commit

 

History

History
293 lines (207 loc) · 10.3 KB

BUILDING-configure.md

File metadata and controls

293 lines (207 loc) · 10.3 KB
Raw

Building using configure

The instructions will assume that you're working from a command-line shell. If you prefer to work from an IDE, you'll have to know how your IDE likes to do things, and you'll want to follow the shell instructions as a guide.

Prerequisites

Before building and installing the libpqxx library, you first need to install the libpq library. This is the official C client library for PostgreSQL that allows libpqxx to communicate with the database backend server. Methods for installing the libpq library will vary depending on your operating system and individual setup.

Quick start

Once you have downloaded the libpqxx source code, you can build and install it quickly by entering the following commands:

    ./configure --disable-shared
    make
    sudo make install

This installation automatically places the libpqxx libraries in the appropriate locations for your system. If you wish, you may delete the downloaded folder after the installation process completes.

Want more detail? Read on.

Stages

I'll explain the main build steps in more detail below, but here's a quick overview:

  1. Configure
  2. Compile
  3. Test
  4. Install

The Test step is optional.

Configure

The configure script configures your build. It figures out various parameters, such as where libpq and its headers are, which C++ features your compiler supports, and which options your compiler needs. It generates Makefiles, which in turn tell the make utility how to perform tasks such as compiling libpqxx, running tests, cleaning up after itself, and installing libpqxx.

The configure step is also where you can set these options, e.g. to instruct the compiler to look for libpq in a non-standard place, or to use a different compiler, or pass different compiler flags. Don't try to specify those while doing the actual compile; set them once when running configure.

Let's say $BUILD is the directory where you want to build libpqxx, and $SRC is where its source code is. So for example, the readme file will be at $SRC/README.md.

In the simplest case, you just do:

    cd $BUILD
    $SRC/configure

Add configure options as needed. There's more about the options below, or in the output of configure --help. I'll also explain the two directories.

Cheat sheet

Here are some popular configure options:

  • --disable-documentation skips building of the documentation.
  • CXXFLAGS=-O0 disables optimisation. Slower code, but faster build.
  • CXXFLAGS=-O3 asks for more optimisation. Faster code, slower build.
  • CXX=clang++ compiles with clang++ as the compiler.
  • --enable-maintainer-mode makes the compiler more pedantic about the code.
  • --enable-audit enables expensive run-time checks for debugging.
  • --with-postgres-lib=$DIR looks for libpq in $DIR.
  • --with-postgres-include=$DIR looks for the libpq headers in $DIR.
  • --prefix=$PATH prepares to install libpqxx in $PATH.
  • --enable-shared enables compilation of libpqxx as a shared library.
  • --disable-shared disbles compilation of libpqxx as a shared library.
  • --enable-static enables compilation of libpqxx as a static library.
  • --disable-static disables compilation of libpqxx as a static library.
  • --help shows you a lot more of the options.

So for example, to get a very quick build but produce very inefficient code:

    ./configure --disable-documentation CXXFLAGS=-O0

Or if you want to pull out all the stops to find problems in the code:

    ./configure --enable-maintainer-mode --enable-audit CXXFLAGS=-O3

(Requesting -O3 optimisation will make some compilers perform extra analysis which may, as a side effect, cause them to notice and warn about certain kinds of mistakes in the code, such as occasionally-unused variables.)

Finding libpq

One of configure's most important jobs in the libpqxx build is to find the headers and library for libpq. It has three ways of finding those:

  1. Asking a popular tool called pkg-config, if installed.
  2. Asking postgres' deprecated pg_config tool, if installed.
  3. Through explicit command-line options to configure.

The explicit command-line options are --with-postgres-lib (for the libpq library binary) and --with-postgres-include (for the libpq headers).

If you want to use a version of libpq that's not installed in a standard location, e.g. if you're cross-compiling to produce a binary for a different CPU architecture than your native system's, use the explicit options.

Where does the configure script come from?

I didn't write the configure script. It was generated by GNU autoconf and related GNU tools. There's a script to re-generate it, called autogen.sh.

The contents of configure are based on a higher-level script called configure.ac. This is where I script checks for specific features in libpq or the compiler. The configure script adds a lot of built-in items that I don't need to worry about, such as figuring out exactly how your build tools work.

Don't try to debug configure yourself if you can help it. It's very hard to read, partly because it's automatically generated, but also because it is engineered to work with an extremely broad range of shells, compilers, tools, and operating systems. If you're going to do a "deep dive," try looking at configure.ac instead.

Source and Build trees

Where should you run configure?

Two directories matter when building libpqxx: the source tree (where the libpqxx source code is) and the build tree (where you want your build artefacts). Here I will call them $SRC and $BUILD, but you can call them anything you like.

They can be one and the same, if you like. It's convenient, but less clean, as source code and build artefacts will exist in the same directory tree. If you're going to delete the source tree after installing, of course it's fine to make a mess in there.

Compile

To start the compile, run the make tool. It will go through all the steps to produce a libpqxx library binary.

Beware though, it only runs one compiler process at a time. That could take a while. Use the -j option to make it run concurrent processes, e.g.:

    make -j8

Very roughly speaking, it's probably fastest if you run one process per CPU core in your system. If you have the nproc utility installed:

    make -j$(nproc)

If you want a very fast build and don't mind missing out on efficient code or documentation, tweak the Configure step above by adding configure options like CXXFLAGS=-O0 and --disable-documentation.

Test

Of course libpqxx comes with a test suite, to check that the library is functioning correctly.

You can run it, but there's one caveat: you need to give it a database where it can log in, without a password or any other parameters, and try out various things.

And when I say you need to "give" it a database, I really mean "give." The test suite will create and drop tables. Those will all have names prefixed with "pqxx", so it's probably safe to use a database you already had, but if any of the items in your database happen to have names starting with pqxx, tough luck. They're fair game.

Enter this in your shell to build and run the tests:

    make check

As with compiling, use the -j option to make better use of your CPUs. For example:

    make check -j$(nproc)

Configuring the test database

But what if you do need a password to log into your test database? Or, what if it's running on a different system so you need to pass that machine's address? What if it's not running on the default port?

You can set these parameters for the test suite, or for any other libpq-based application, using the following environment variables. (They only set default values, so they won't override parameters that the application sets in some other way.)

  • PGHOST — the IP address where we can contact the database's socket. Or for a Unix domain socket, its absolute path on the filesystem.
  • PGPORT —
  • PGDATABASE — the name of the database to which you wish to connect.
  • PGUSER — user name under which you wish to log in on the database.
  • PGPASSWORD — user name's password for accessing the database.

See the full list at https://www.postgresql.org/docs/current/libpq-envars.html .

Be careful with passwords, by the way. Depending on your operating system and configuration, an attacker with access to your machine could try to read your password if you set it on the command line:

  • Your shell may keep a log of the commands you have entered.
  • Environment variables may be visible to other users on the system.

If at all possible, rely on postgres "peer authentication." Once set up, it is both more secure and more convenient than passwords.

Install

Installing libpqxx will install the library and headers in a location chosen at the time you can the configure script. On some systems it defaults to the /usr/local/ tree, but it may be different in your environment. Or, use the configure script's --prefix option to set an install location.

(If you want to see exactly what happens, you can run any make command line with the -n option, which means: don't actually do this, but print all the commands you would execute if you did. It's a lot of output though.)

To install, ensure that you have sufficient privileges to write the files to their install locations, and run:

    make install

Save your build tree somewhere, so that you will be able to undo installation in the future:

    make uninstall

When using the library, make sure the libpqxx headers are in your compiler's include path. (You will no longer need the libpq headers at that time.)

Also, building an application which uses libpqxx, make sure the libpqxx library binary is in your compiler's library search path. And if the library binary is a shared library, you'll also need it in your loader's search path when running your application.

This last part goes for libpq as well: when using libpq, make sure you have the libpq library binary in your compiler's library search path, and if it's a shared library, also have it in your loader's search path when running.