I was installing Anaconda Python on Linux. For Linux, Anaconda provide a bash script, but the file is huge, almost 300 MB. I decided to see why, and opened it in a text editor.

About 95% of the file is machine language gibberish, like this:

ºîØôЕzÒA¶©h¶¥R•„&´ìÒUÓçß3{^eÑòà(|ÄÃk뎆ºîØôЕzÒA¶©h¶¥R•„&´ìÒUÓçß3{™½ö|q ŽÖm¶¥¡ôÚ­gú¡@óìÛkkº£C»Iš)à÷¾Û¸êw½æõîJN7í×p€A¡ÈzÞÝï8

The file isn't corrupt, as I can install Python.

Most of this is in the license file, so Im wondering if it's unicode for another language, but that wouldn't take 95% of the file, would it?

Can it be compiled code / machine language? Is it allowed to put machine code in bash files?

  • 4
    It's probably a self extracting binary. The actual script part will copy the gibberish part to a separate file(s), most likely doing some decompression along the way, then execute it. As long as the scrip exits before it reaches the non interpretable part, this is completely fine. – Graeme Sep 26 '15 at 9:43
  • bash can handle embedded data in binary – Skaperen Sep 26 '15 at 9:45
  • 4
    For future reference, machine language or machine code is only one kind of binary data. Your file definitely has binary data in it, but from what you've shown we don't know if it's machine code or not. – zwol Sep 26 '15 at 15:16

To expand on @Graeme's comment.

The downloaded script is a bash script with an embedded tarball. The script part first validates the tarball by md5sum, then unpack the tar, which contains multiple .tar.bz2 archives. Then it proceeds by using a custom function extract_dist() to unpack the archives. I.e.:

extract_dist python-2.7.10-0
extract_dist conda-3.14.1-py27_0

which extracts the files:


For the 32-bit version the script part can be extracted by:

head -n 467 Anaconda-2.3.0-Linux-x86.sh

For the 64-bit version the script part can be extracted by:

head -n 466 Anaconda-2.3.0-Linux-x86_64.sh

As you can see the script part ends with exit 0 which aborts any further processing of the script by bash.

The tarball is extracted by:

tail -n +469 $THIS_PATH | tar xf - --no-same-owner
tail -n +468 $THIS_PATH | tar xf - --no-same-owner

for 32-bit and 64-bit respectively.

You could for example do:

tail -n +469 Anaconda-2.3.0-Linux-x86.sh | tar -t

to list the files in the 32-bit archive.

  • Thanks. Is there a reason to do it this way, rather than just include a tar ball with the bash script, and extract it that way? – Shantnu Tiwari Sep 27 '15 at 9:52
  • 1
    @ShantnuTiwari: For one, limit the n factors of faults that can happen between the keyboard and the chair. User downloads the 32-bit script and the 64-bit tar: Error. User downloads the 64-bit-script and the 32-bit tar: Error. User downloads old version of script and new tar: Error. Those are some of the most simple examples. Ease of distribution: distribute one file over e.g. torrent, FTP, HTTP or what ever. You could pack a compressed archive holding the install script and the tarball from where end user is expected to extract, read a README on "how to" etc. but ... – Runium Oct 3 '15 at 17:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.