Editing 1654: Universal Install Script

Most users of computers today are used to simple, easy installation of programs. You just download a .exe or a .pkg, double click it, and do what it says. Sometimes you don't even have to install anything at all, and it runs without any installation.

However, when things are more "homebrew", for example downloading source code, things are more complicated. Under {{w|Unix-like}} systems, which this universal install script is designed for, you may have to work with "build environments" and "makefiles", and command line tools. To make this process simpler, there exist repositories of programs which host either packages of source code and the things needed to build it or the pre-built programs. When you download the package, it automatically does most of the work of building the code into something executable if necessary and then installing it. However, there are many such repositories, such as "pip" and "brew", among others listed in the comic. If you only know the name of a program or package, you may not know in which repository(ies) it resides.

 

The script accepts the name of a program when you run it as an argument. This value is then referenced as "$1" (argument number 1). Everywhere the script says "$1", it substitutes in the name of the package you gave it. The end result is the name being tried against a large number of software repositories and package managers, and hopefully, at least one of them will be appropriate and the program will be successfully installed. Near the end, it even tries changing the current working directory to that which is assumed to hold the package to be installed, and then runs several commands which build the program from source code.

One of the more subtle jokes in the comic is the inclusion of <code>apt-get</code> and <code>sudo apt-get</code> in the same script. In most cases this would be redundant as the <code>sudo</code> command is just to add admin permissions. This could be an allusion to a joke in the Linux community about forgetting to include the sudo command. An example of this joke being used elsewhere was a  [https://twitter.com/liamosaur/status/506975850596536320 viral tweet] that showed a workaround for the issue. Sudo has also been used both by [[Randall]] in [[149: Sandwich]] and by Jason Fox to force Randall to let him appear on xkcd with [[824: Guest Week: Bill Amend (FoxTrot)]].

Another explanation for this could be that plain "apt-get" is for Debian, while Ubuntu etc. use sudo.

The tool <code>curl</code> downloads files from the network (e.g., the Internet). Used like <code>curl http://xkcd.com/</code> it downloads the xkcd main page and displays the HTML source code. The pipe <code>|</code> in the script attaches the output of the command before the pipe to the input of the command after the pipe. Both commands are executed concurrently. Bash is a popular shell for Unix-like operating systems. The line <code>curl "$1" | bash</code> tries to download a file from the network and to execute the download directly. <!--- Although this is a common practice for conveniently installing software, it is considered extremely insecure and should never be done. [ed. note: there's no reason this cannot be secure, especially if HTTPS with validated certificates is used, from a trusted domain which utilizes DNSSEC] -->

The use of & at the end of each line causes the shell interpreter to execute the commands in parallel (asynchronously) instead of sequentially. Even if single commands fail, the rest of them will be executed. Note this is even the case for the final commands which attempt to change to the installed package, probably the only reason why this may not work completely for packages that do need compiling after being downloaded. (However, just running this script again would probably do the trick.)

There appears to be a bug with the & at the end of the "git clone" line; since a git repository typically contains program source code, not executables, it may have been intended to retrieve the source code with git and then compile and install the program in the next line. In this case, the single & should be replaced with &&, an operator that will run the second command only if the first one has completed successfully. This plays into a second bug on the "configure" line, where the placement of the & means that only the "make install" command will be run asynchronously after the "configure" and "make" steps have finished in sequence. To make success as likely as possible, the two lines should be like this:

git clone <nowiki>https://github.com/</nowiki>"$1"/"$1" && (cd "$1"; ./configure; make; make install) &

The title text mentions the possibility that the same program may be in multiple repositories, so in this case, the script will download and install several versions, or it may fail on a number of repositories, in which case usually nothing bad happens. Since all the commands come from different operating systems, versions, or distributions, it is not very likely that more than one will work (with the exception of pip/easy_install and the two forms of apt-get) or even exist on the same system. It mentions that adding a way of automatically saying "yes" to questions asked during the different repository-fetching programs' running, by making them read input from another program that writes a (nearly) endless stream of "y"s, could simplify things further. <code>2>/dev/null</code> redirects the second output stream (the "error stream") to the null device driver, which discards all writes to it, meaning errors (the package not existing) will be ignored.

*steamcmd refers to {{w|Stream (software)|Steam}}, the computer game client