YAML, Docker Compose, Spaces & Quotes

Question

Under what circumstances must one use quotes in a YAML file, specifically when using docker-compose.

For instance,

service:   image: "my-registry/repo:tag1"   environment:     ENV1: abc     ENV2: "abc"     ENV3: "a b c" 

If spaces are required, for example, must one use quotes around the environment variable, as depicted in ENV3?

Answer 1

After some googling I've found a blog post that touches this problem as I understood it.

I'll cite the most important part here:

plain scalars: - a string - a string with a \ backslash that doesn't need to be escaped - can also use " quotes ' and $ a % lot /&?+ of other {} [] stuff  single quoted: - '& starts with a special character, needs quotes' - 'this \ backslash also does not need to be escaped' - 'just like the " double quote' - 'to express one single quote, use '' two of them'  double quoted: - "here we can use predefined escape sequences like \t \n \b" - "or generic escape sequences \x0b \u0041 \U00000041" - "the double quote \" needs to be escaped" - "just like the \\ backslash" - "the single quote ' and other characters must not be escaped"  literal block scalar: |   a multiline text   line 2   line 3  folded block scalar: >   a long line split into   several short   lines for readability 

Also I have not seen such docker-compose syntax to set env variables. Documentation suggests using simple values like

environment:   - ENV1=abc   - "ENV2=abc" 

Where quotes " or ' are optional in this particular example according to what I've said earlier.

To see how to include spaces in env variables you can check out this so answer

Answer 2

Whether or not you need quotes, depends on the parser. Docker-compose AFAIK is still relying on the PyYAML module and that implements most of YAML 1.1 and has a few quirks of its own.

In general you only need to quote what could otherwise be misinterpreted or clash with some YAML construct that is not a scalar string. You also need (double) quotes for things that cannot be represented in plain scalars, single quoted scalars or block style literal or folded scalars.

Misinterpretation

You need to quote strings that look like some of the other data structures:

• booleans: "True", "False", but PyYAML also assumes alternatives words like "Yes", "No", "On", "Off" represent boolean values ( and the all lowercase, all uppercase versions should be considered as well). Please note that the YAML 1.2 standard removed references to these alternatives.
• integers: this includes string consisting of numbers only. But also hex (0x123) and octal number (0123). The octals in YAML 1.2 are written as 0o123, but PyYAML doesn't support this, however it is best to quote both. A special integer that PyYAML still supports but again not in the YAML 1.2 specification are sexagesimals: base 60 number separated by colon (:), time indications, but also MAC addresses can be interpreted as such if the values between/after the colons are in the range 00-59
• floats: strings like 1E3 (with optional sign ans mantissa) should be quoted. Of course 3.14 needs to be quoted as well if it is a string. And sexagesimal floats (with a mantissa after the number after the final colon) should be quoted as well.
• timestamps: 2001-12-15T02:59:43.1Z but also iso-8601 like strings should be quoted to prevent them from being interpreted as timestamps
• The null value is written as the empty string, as ~ or Null (in all casing types), so any strings matching those need to be quoted.

Quoting in the above can be done with either single or double quotes, or block style literal or folded scalars can be used. Please note that for the block-style you should use |- resp. >- in order not to introduce a trailing newline that is not in the original string.

Clashes

YAML assigns special meaning to certain characters or character combinations. Some of these only have special meaning at the beginning of a string, others only within a string.

• characters fromt the set !&*?{[ normally indicate special YAML constructs. Some of these might be disambiguated depending on the following character, but I would not rely on that.
• whitespace followed by # indicates an end of line comment
• wherever a key is possible (and within block mode that is in many places) the combination of colon + space (:) indicates a value will be following. If that combination is part of your scalar string, you have to quote.

As with the misinterpretation you can use single or double quoting or block-style literal or folding scalars. There can be no end-of-line comments beyond the first line of a block-style scalar.

PyYAML can additionally get confused by any colon + space within a plain scalar (even when this is in a value) so always quote those.

Representing special characters

You can insert special characters or unicode code-points in a YAML file, but if you want these to be clearly visible in all cases, you might want to use escape sequences. In that case you have to use double quotes, this is the only mode that allows backslash escapes. And e.g. \u2029. A full list of such escapes can be taken from the standard, but note that PyYAML doesn't implement e.g \/ (or at least did not when I forked that library).

---

One trick to find out what to quote or not is to use the library used to dump the strings that you have. My ruamel.yaml and PyYAML used by docker-compose, when potentially dumping a plain scalar, both try to read back (yes, by parsing the result) the plain scalar representation of a string and if that results in something different than a string, it is clear quotes need to be applied. You can do so too: when in doubt write a small program dumping the list of strings that you have using PyYAML's safe_dump() and apply quotes anywhere that PyYAML does.