main
Raw Download raw file
  1// Package gcfg reads "INI-style" text-based configuration files with
  2// "name=value" pairs grouped into sections (gcfg files).
  3//
  4// This package is still a work in progress; see the sections below for planned
  5// changes.
  6//
  7// Syntax
  8//
  9// The syntax is based on that used by git config:
 10// http://git-scm.com/docs/git-config#_syntax .
 11// There are some (planned) differences compared to the git config format:
 12//  - improve data portability:
 13//    - must be encoded in UTF-8 (for now) and must not contain the 0 byte
 14//    - include and "path" type is not supported
 15//      (path type may be implementable as a user-defined type)
 16//  - internationalization
 17//    - section and variable names can contain unicode letters, unicode digits
 18//      (as defined in http://golang.org/ref/spec#Characters ) and hyphens
 19//      (U+002D), starting with a unicode letter
 20//  - disallow potentially ambiguous or misleading definitions:
 21//    - `[sec.sub]` format is not allowed (deprecated in gitconfig)
 22//    - `[sec ""]` is not allowed
 23//      - use `[sec]` for section name "sec" and empty subsection name
 24//    - (planned) within a single file, definitions must be contiguous for each:
 25//      - section: '[secA]' -> '[secB]' -> '[secA]' is an error
 26//      - subsection: '[sec "A"]' -> '[sec "B"]' -> '[sec "A"]' is an error
 27//      - multivalued variable: 'multi=a' -> 'other=x' -> 'multi=b' is an error
 28//
 29// Data structure
 30//
 31// The functions in this package read values into a user-defined struct.
 32// Each section corresponds to a struct field in the config struct, and each
 33// variable in a section corresponds to a data field in the section struct.
 34// The mapping of each section or variable name to fields is done either based
 35// on the "gcfg" struct tag or by matching the name of the section or variable,
 36// ignoring case. In the latter case, hyphens '-' in section and variable names
 37// correspond to underscores '_' in field names.
 38// Fields must be exported; to use a section or variable name starting with a
 39// letter that is neither upper- or lower-case, prefix the field name with 'X'.
 40// (See https://code.google.com/p/go/issues/detail?id=5763#c4 .)
 41//
 42// For sections with subsections, the corresponding field in config must be a
 43// map, rather than a struct, with string keys and pointer-to-struct values.
 44// Values for subsection variables are stored in the map with the subsection
 45// name used as the map key.
 46// (Note that unlike section and variable names, subsection names are case
 47// sensitive.)
 48// When using a map, and there is a section with the same section name but
 49// without a subsection name, its values are stored with the empty string used
 50// as the key.
 51// It is possible to provide default values for subsections in the section
 52// "default-<sectionname>" (or by setting values in the corresponding struct
 53// field "Default_<sectionname>").
 54//
 55// The functions in this package panic if config is not a pointer to a struct,
 56// or when a field is not of a suitable type (either a struct or a map with
 57// string keys and pointer-to-struct values).
 58//
 59// Parsing of values
 60//
 61// The section structs in the config struct may contain single-valued or
 62// multi-valued variables. Variables of unnamed slice type (that is, a type
 63// starting with `[]`) are treated as multi-value; all others (including named
 64// slice types) are treated as single-valued variables.
 65//
 66// Single-valued variables are handled based on the type as follows.
 67// Unnamed pointer types (that is, types starting with `*`) are dereferenced,
 68// and if necessary, a new instance is allocated.
 69//
 70// For types implementing the encoding.TextUnmarshaler interface, the
 71// UnmarshalText method is used to set the value. Implementing this method is
 72// the recommended way for parsing user-defined types.
 73//
 74// For fields of string kind, the value string is assigned to the field, after
 75// unquoting and unescaping as needed.
 76// For fields of bool kind, the field is set to true if the value is "true",
 77// "yes", "on" or "1", and set to false if the value is "false", "no", "off" or
 78// "0", ignoring case. In addition, single-valued bool fields can be specified
 79// with a "blank" value (variable name without equals sign and value); in such
 80// case the value is set to true.
 81//
 82// Predefined integer types [u]int(|8|16|32|64) and big.Int are parsed as
 83// decimal or hexadecimal (if having '0x' prefix). (This is to prevent
 84// unintuitively handling zero-padded numbers as octal.) Other types having
 85// [u]int* as the underlying type, such as os.FileMode and uintptr allow
 86// decimal, hexadecimal, or octal values.
 87// Parsing mode for integer types can be overridden using the struct tag option
 88// ",int=mode" where mode is a combination of the 'd', 'h', and 'o' characters
 89// (each standing for decimal, hexadecimal, and octal, respectively.)
 90//
 91// All other types are parsed using fmt.Sscanf with the "%v" verb.
 92//
 93// For multi-valued variables, each individual value is parsed as above and
 94// appended to the slice. If the first value is specified as a "blank" value
 95// (variable name without equals sign and value), a new slice is allocated;
 96// that is any values previously set in the slice will be ignored.
 97//
 98// The types subpackage for provides helpers for parsing "enum-like" and integer
 99// types.
100//
101// Error handling
102//
103// There are 3 types of errors:
104//
105//  - programmer errors / panics:
106//    - invalid configuration structure
107//  - data errors:
108//    - fatal errors:
109//      - invalid configuration syntax
110//    - warnings:
111//      - data that doesn't belong to any part of the config structure
112//
113// Programmer errors trigger panics. These are should be fixed by the programmer
114// before releasing code that uses gcfg.
115//
116// Data errors cause gcfg to return a non-nil error value. This includes the
117// case when there are extra unknown key-value definitions in the configuration
118// data (extra data).
119// However, in some occasions it is desirable to be able to proceed in
120// situations when the only data error is that of extra data.
121// These errors are handled at a different (warning) priority and can be
122// filtered out programmatically. To ignore extra data warnings, wrap the
123// gcfg.Read*Into invocation into a call to gcfg.FatalOnly.
124//
125// TODO
126//
127// The following is a list of changes under consideration:
128//  - documentation
129//    - self-contained syntax documentation
130//    - more practical examples
131//    - move TODOs to issue tracker (eventually)
132//  - syntax
133//    - reconsider valid escape sequences
134//      (gitconfig doesn't support \r in value, \t in subsection name, etc.)
135//  - reading / parsing gcfg files
136//    - define internal representation structure
137//    - support multiple inputs (readers, strings, files)
138//    - support declaring encoding (?)
139//    - support varying fields sets for subsections (?)
140//  - writing gcfg files
141//  - error handling
142//    - make error context accessible programmatically?
143//    - limit input size?
144//
145package gcfg // import "github.com/go-git/gcfg"