|
libconfini
Yet another INI parser
|
IniFormat data type
IniFormatNum data typeIniStatistics and IniDispatch structuresUser Guide for libconfini
libconfini is a simple INI parsing library with the ability to read disabled entries (i.e. valid entries nested in comments). libconfini does not store the data read from an INI file, but rather dispatches it, formatted, to a custom listener.
The code is written in C (C99) and does not depend on any particular library, except for the C standard headers stdio.h, stdlib.h, stdbool.h and stdint.h.
If you want to start learning directly from the code, you can find partially self-documented sample usages of libconfini under /usr/share/doc/libconfini/examples.
INI files were introduced with the early versions of Microsoft Windows, where the .ini file name extension stood for INItialization. An INI file can be considered as a string representation of a tree object, with new lines used as delimiters between nodes. A typical INI file is a plain text file looking like the following example:
During the years several interpretations of INI files appeared. In some implementations the colon character (:) was adopted as delimiter between keys and values instead of the classic equals sign (a typical example under GNU is /etc/nsswitch.conf); in other implementations, under the influence of Unix standard configuration files, a sequence of one or more spaces (/[ \t\v\f]+/ or /(?:\\(?:\n\r?|\r\n?)|[\t \v\f])+/) was adopted instead (see for example /etc/host.conf).
Equals sign used as delimiter between keys and values:
Colon sign used as delimiter between keys and values:
Space sequence used as delimiter between keys and values:
libconfini was born as a general INI parser for GNU, so the support of most part of INI dialects has been implemented within it.
Especially in Microsoft Windows a more radical syntax variation found its way into INI files: the use of semicolons, instead of new lines, as delimiter between nodes, as in the following example:
For several reasons the use of semicolons as node delimiters is not (and will never be) supported by libconfini.
A key-value element is identified as a string placed after a new line and followed by a key-value delimiter – typically the equals sign (=) or the colon sign (:) or a space sequence (/\s+/) – which is followed by a value, which is followed by a new line or an inline comment.
Both the key part and the value part may be enclosed within quotes (' or "):
The key part can contain any character, except the delimiter (although this may be enclosed within quotes for not beeing considered as such). In multi-line formats internal new line sequences must be escaped (/\\(?:\n\r?|\r\n?)/).
If the key part part is missing libconfini considers the element of unknown type (example: = foo). If the value part is missing the key element is considered empty (example: foo =). If the delimiter is missing (and therefore the value part as well), according to some formats the key element is is considered to be an implicit key – typically representing the boolean true (example: foo). For instance, in the following example from /etc/pacman.conf, IgnorePkg is an empty key, while Color is an implicit key representing a true boolean – i.e. Color = YES:
The value part can contain typed data, usually: a boolean (booleans supported by libconfini are: FALSE/TRUE, NO/YES, OFF/ON – case-insensitive), a string, a number, or an array (typically with commas or spaces as delimiters between members – examples: paths = /etc, /usr, /home/john/Personal Data or paths = /etc /usr "/home/john/Personal Data"). In multi-line formats internal new line sequences must be escaped (/\\(?:\n\r?|\r\n?)/).
A section can be imagined as a directory. A section path is identified as the string "$1" in the regular expression /(?:^|\n|\r)[ \t\v\f]*\[[ \t\v\f]*([^\]]*)[ \t\v\f]*\][ \t\v\f]*(?:\n|\r|$)/ globally applied to an INI file. A section path expresses nesting using the “dot” character, as in the following example:
A section path starting with a dot expresses nesting to the previous section. Hence, the last example is equivalent to:
Keys appearing before any section path belong to a virtual root node (with an empty string as path), as the key foo in the following example:
Section parts may be enclosed within quotes:
Comments are string segments enclosed within the sequence /(?:^|\s)[;#]/ and a new line sequence, as in the following example:
Comments may in theory be multi-line, following the same syntax of multi-line disabled entries (see below). This is usually of little utility, except for inline comments you want to make sure will refer to the previous entry:
A disabled entry is either a section or a key that has been nested inside a comment as its only child. Inline comments cannot represent disabled entries. According to some formats disabled entries can be multi-line, using /\\(?:\n\r?|\r\n?)[\t \v\f]*[;#]/ as multi-line escape sequence. For example:
In order to maximize the flexibility of the data, only four escape sequences are supported by libconfini: \', \", \\ and the multi-line escape sequence (/\\(?:\n\r?|\r\n?)/).
The first three escape sequences are left untouched by all functions except ini_string_parse() and ini_unquote() (see below). Nevertheless, the characters ', " and \ can determine different behaviors during the parsing depending on whether they are escaped or unescaped. For instance, the string johnsmith !" in the following example will not be parsed as a comment:
A particular case of escape sequence is the multi-line escape sequence (/\\(?:\n\r?|\r\n?)/), which in multi-line INI files gets immediately unescaped by libconfini.
The API is contained in one single public header:
When libconfini is used as a shared library, it might be wiser to include the versioned header (with only the major version number appended to the file name), in order to ensure that the code will compile correctly even when different major versions of the library cohabit in the same machine. This can apply also to version 1.X.X:
1. Using a pointer to a FILE handle:
2. Using a path:
Where:
ini_file in load_ini_file() is the FILE handle pointing to the INI filepath in load_ini_path() is the path where the INI file is located (pointer to a char array, a.k.a. a "C string")format is a bitfield that defines the syntax of the INI file (see struct IniFormat)f_init is the function that will be invoked before any dispatching begins – it can be NULLf_foreach is the callback function that will be repeatedly invoked for each member of the INI file - it can be NULLuser_data is a pointer to a custom argument – it can be NULLThe user given function f_init (see IniStatsHandler data type) will be invoked with two arguments:
statistics – a pointer to an IniStatistics object containing some properties about the file read (like its size in bytes and the number of its members)user_data – a pointer to the custom argument previously passed to the load_ini_file() / load_ini_path() functionsThe user given function f_foreach (see IniDispHandler data type) will be invoked with two arguments:
dispatch – a pointer to an IniDispatch object containing the parsed member of the INI fileuser_data – a pointer to the custom argument previously passed to the load_ini_file() / load_ini_path() functionsBoth functions load_ini_file() and load_ini_path() will return zero if the INI file has been completely dispatched, non-zero otherwise.
Example 1. Using a pointer to a FILE handle:
Example 2. Using a path:
The function load_ini_path() is a clone of load_ini_file() that requires a path instead of a FILE handle.
Both functions load_ini_file() and load_ini_path() dynamically allocate at once the whole INI file into the heap, and the two structures IniStatistics and IniDispatch into the stack. All members of the INI file are then dispatched to the custom listener f_foreach(). Finally, the allocated memory gets automatically freed.
Because of this mechanism it is very important that all the dispatched data be immediately copied by the user (when needed), and no pointers to the passed data be saved: after the end of the functions load_ini_file() / load_ini_path() all the allocated data will be destroyed indeed – and moreover each dispatch might overwrite data from previous dispatches.
Within a dispatching cycle, the structure containing each dispatch (IniDispatch * dispatch) is always the same struct that gets constantly updated with new information.
Note: On some platforms, such as Microsoft Windows, it might be necessary to add the binary specifier ("b") to the mode string of the FILE handle passed to load_ini_file() in order to prevent discrepancies between the physical size and the computed size of the file. Adding the binary specifier guarantees portability across all platforms:
Starting from version 1.10.0, it is possible to parse a disposable buffer containing an INI file instead of a physical file (i.e. to parse a char array). The function that allows to do so is named strip_ini_cache(). This function presents some important differences compared to load_ini_file() and load_ini_path():
strip_ini_cache() – you can use strndup() for this, or use the example belowload_ini_file() and load_ini_path(), every needed information must be copied immediately with each dispatchIn short, strip_ini_cache() works exactly like load_ini_file() and load_ini_path(), but with the difference that it destroys the input while it dispatches it. And of course the input is not anymore a file, but a disposable buffer instead. As a matter of fact, strip_ini_cache() is the main parsing function both load_ini_file() and load_ini_path() rely on in order to dispatch the content of an INI file. For a sample usage, please see examples/topics/strip_ini_cache.c.
If you want to automatize the process of making a copy of a read-only buffer, strip and parse the copy, then free the allocated memory, you can use the following function:
The function above can then be invoked directly on a const buffer:
Since in most cases an INI buffer is a disposable buffer (unless one wants to parse the very same buffer more than once), libconfini's interface does not include the function in the example above.
IniFormat data typeFor a correct use of this library it is helpful to understand the IniFormat data type. libconfini has been born as a general INI parser, with the main purpose of being able to parse INI files written by other programs (see Rationale), therefore some flexibility was required. When an INI file is parsed it is parsed according to a particular format. The IniFormat data type is a univocal description of such format. It is implemented as a 24-bit bitfield. Its small size (3 bytes) allows it to be passed by value to the functions that require it.
Since no function requires a pointer to an IniFormat data type as argument, a preprocessor macro can be a good place where to store a custom format:
A default format named INI_DEFAULT_FORMAT is available.
The code above corresponds to:
or, equivalently, in macro form:
Starting from version 1.7.0 a format named INI_UNIXLIKE_FORMAT is available.
This format is a clone of INI_DEFAULT_FORMAT with the only exception of IniFormat::delimiter_symbol, whose value is set to INI_ANY_SPACE instead of INI_EQUALS.
The semantics of the IniFormat bitfield has been designed to ensure that when all its fields are set to zero it equals INI_UNIXLIKE_FORMAT.
For further formats, please refer to the extensible list of common formats in the project's wiki. Feel free to contribute.
IniFormatNum data typeEach format can be represented also as a univocal 24-bit unsigned integer. In order to convert an IniFormat to an unsigned integer and vice versa the functions ini_fton() and ini_ntof() are available. For simplicity, instead of using a uint_least32_t type, a size-agnostic custom type is used for this: the IniFormatNum data type.
For instance, imagine we want to create a format as close as possible to the typical Windows INI files. Probably we would define our format as follows:
According to the ini_fton() function this format is univocally the format No. 56637. The function ini_ntof() then gives us a shortcut to construct the very same format using its format number. Hence, the code above corresponds to:
IniStatistics and IniDispatch structuresWhen the functions load_ini_file(), load_ini_path() read an INI file, or when the function strip_ini_cache() parses a buffer, they dispatch the file content to the f_foreach() listener. Before the dispatching begins some statistics about the parsed file can be dispatched to the f_init() listener (if this is non-NULL).
The information passed to f_init() is passed through an IniStatistics structure, while the information passed to f_foreach() is passed through an IniDispatch structure.
The output strings dispatched by libconfini follow some formatting rules depending on their role within the INI file. First, multi-line escape sequences will be unescaped, then
key_name.replace(/^[\n\r]\s*|\s+/g, " ") – within single or double quotes, if active, the text will be rendered verbatim.section_name.replace(/\.*\s*$|(?:\s*(\.))+\s*|^\s+/g, "$1").replace(/\s+/g, " ") – within single or double quotes, if active, the text will be rendered verbatim – otherwise, will be rendered according to the same algorithm used for key names.format.do_not_collapse_values is active, will only be cleaned of spaces at the beginning and at the end; otherwise will be rendered according to the same algorithm used for key names (with the difference that, if format.preserve_empty_quotes is set to true, empty quotes surrounded by spaces will be preserved).comment_string.replace(/(^|\n\r?|\r\n?)[ \t\v\f]*[#;]+/g, "$1"); elsewhere, according to ECMAScript comment_string.replace(/^[ \t\v\f]*[#;]+/, "").In order to perform comparisons between the strings dispatched the functions ini_string_match_ss(), ini_string_match_si(), ini_string_match_ii() and ini_array_match() are available. The function ini_string_match_ss() compares two simple strings, the function ini_string_match_si() compares a simple string with an unparsed INI string, the function ini_string_match_ii() compares two unparsed INI strings, and the function ini_array_match() compares two INI arrays. INI strings are the strings typically dispatched by load_ini_file(), load_ini_path() or strip_ini_cache() which may contain quotes and the three escape sequences \\, \', \". Simple strings are user-given strings or the result of ini_string_parse().
As a consequence, the functions ini_string_match_si(), ini_string_match_ii() and ini_array_match() do not perform literal comparisons of equality between strings. For example, in the following (absurd) INI file the two keys foo and hello belong to the same section named this is a double quotation mark: "! (after being parsed by ini_string_parse()).
Therefore...
Or, for instance, in the following example the first two arrays are considered equal, while the third one is considered different.
In formats that support quotes, the function ini_array_match() is also the function that should be used, with '.', or INI_DOT, as delimiter (see enum IniDelimiters), to compare properly section paths containing more than one level of nesting.
In case of multiple comparisons you might want to use a macro:
The four functions ini_string_match_ss(), ini_string_match_si(), ini_string_match_ii(), ini_array_match() perform case-sensitive or case-insensitive comparisons depending on the format given. The comparison between UTF-8 codepoints out of the ASCII range is always case-sensitive.
Note that within INI strings empty quotes and spaces out of quotes are always collapsed during comparisons. Furthermore, remember that the multi-line escape sequence /\\(?:\n\r?|\r\n?)/ is not considered as such in INI strings, as this is the only escape sequence automatically unescaped by libconfini before each dispatch.
It is possible with libconfini to parse INI values as arrays. In order to avoid that any particular character be treated as a metacharacter throughout an entire configuration file, INI arrays do not have a fixed delimiter symbol specified in the INI format.
Abstractly speaking, this means that array delimiters are not part of the INI syntax, but of the INI semantics instead. Concretely speaking, this means that developers have to provide a delimiter each time they decide to parse an INI string as an array.
Out in the wild, the most widespread INI array delimiter is probably the space sequence (INI_ANY_SPACE). Another widespread delimiter is the comma (',' or INI_COMMA). Besides these two, other characters can occasionally be found as array delimiters too (':', '|', ';', etc.).
For iterating through an INI array using a delimiter, libconfini provides the following functions:
ini_array_get_length()ini_array_foreach()ini_array_collapse()ini_array_break()ini_array_release()ini_array_shift()ini_array_split()None of the functions above actually ever allocates any array, as libconfini does not know what kind of data type an array should be composed of – the developer might be looking for an array of strings, or an array of integers, or anything else. These tools only provide safe mechanisms to tokenize or iterate through the members of INI arrays according to the format and the delimiter given – making sure, for example, that when the delimiter symbol is found nested within quotes it is treated as a normal character, and so on – but allocating memory is something that must be done manually.
An example function that parses INI strings as newly allocated arrays of C strings is available under examples/utilities/make_strarray.h. If your program requires other data types (such as integers, booleans, etc.) you may adapt that example to your needs.
The strings dispatched, as already said, must not be freed. Nevertheless, before being copied or analyzed they can be edited, with some precautions:
IniDispatch::d_len and IniDispatch::v_len).IniDispatch::data field of a section: the IniDispatch::append_to properties of its children will either share this buffer or will concatenate it to another buffer. If you edit its content – and depending on how you edit it – you might be no more able to rely on the IniDispatch::append_to properties of this node's children (you would not make any damage, the loop will continue just fine – if you think you are never going to use the property IniDispatch::append_to, just do it; alternatively, use strndup()).IniDispatch::data contains a key name or a comment, it is guaranteed that no other dispatch will share this buffer, so feel free to edit it before it gets lost.IniDispatch::value, if it does not represent an implicit value (see § Implicit keys) or if IniFormat::implicit_is_not_empty is set to false, this buffer is never shared between dispatches, so feel free to edit it.IniDispatch::append_to is likely to be shared with other dispatches. Again, you would not destroy the world nor generate errors, but you would make the next IniDispatch::append_tos useless. For this reason the buffer pointed by IniDispatch::append_to is passed as constant. To unquote the path parts listed in this field please use strndup().IniDispatch::data, IniDispatch::value and IniDispatch::append_to and the fields IniDispatch::type, IniDispatch::d_len, IniDispatch::v_len and IniDispatch::at_len, are constantly reset, so feel free to use them as custom placeholders if you like – but check their types: IniDispatch::type has only eight bits available and IniDispatch::append_to will always point to a const buffer.IniDispatch::dispatch_id will be checked right before the next dispatch, so you should not edit it. The reading happens only as further diagnostics: if you constantly set this field to 0 the loop will still end at the right moment (hopefully), but if you set it instead to a value equal or higher than the IniStatistics::members previously received, the loop will immediately stop and a CONFINI_EOOR will be thrown (see enum ConfiniInterruptNo). You can think of this as a dirty way to write return !0 at the end of your listener. Unless you really know what you are doing, do not edit IniDispatch::dispatch_id.Typical peaceful edits are the ones obtained by calling the functions ini_array_collapse() and ini_string_parse() directly on the buffer IniDispatch::value:
If all these rules, although thoroughly exposed, still sound confusing to you, always use strndup() on the strings dispatched and feel free to edit your own buffers as you wish. Under examples/utilities/clone_ini_dispatch.h you can find a function designed to make a hard copy of an entire IniDispatch, including all the strings that this points to.
Once your listener starts to receive the parsed data you may want to format and better parse the value part of key-value elements. The following functions might be useful for this purpose:
ini_get_bool()ini_get_bool_i()ini_string_parse()ini_array_get_length()ini_array_foreach()ini_array_collapse()ini_array_break()ini_array_release()ini_array_shift()ini_array_split()Together with the functions listed above the following links are available, in case you don't have #include <stdlib.h> in your source:
Further useful resources include:
examples/utilities/clone_ini_dispatch.hexamples/utilities/make_strarray.hThe function ini_unquote() can be useful for key names enclosed within quotes. This function is very similar to ini_string_parse(), except that does not bother collapsing the sequences of more than one space that might result from removing empty quotes – this is never necessary, since empty quotes surrounded by spaces in key and section names are always automatically collapsed before being dispatched.
You could use ini_string_parse() as well to parse key and section names, but you would obtain the same result with a slightly bigger effort from the CPU.
In order to retrieve the parts of a section path, the functions ini_array_get_length(), ini_array_foreach(), ini_array_break(), ini_array_release(), ini_array_shift() and ini_array_split() can be used with '.' or INI_DOT as delimiter (see enum IniDelimiters). Note that section paths dispatched by libconfini are always collapsed arrays, therefore calling the function ini_array_collapse() on them will have no effects.
It might be required that the function ini_unquote() be applied to each part of a section path, depending on the content and the format of the INI file.
In order to set the value to assign to implicit keys (i.e. keys without a delimiter and a value), please use the ini_global_set_implicit_value() function. A true boolean is usually a good choice:
Alternatively, instead of ini_global_set_implicit_value() you can manually declare at the beginning of your code the two global variables INI_GLOBAL_IMPLICIT_VALUE and INI_GLOBAL_IMPLICIT_V_LEN, which will be retrieved by libconfini:
Or you can assign a value to them at the beginning of the main() function of your program:
If not defined elsewhere, these variables are initialized respectively to NULL and 0 by default.
Although the two variables INI_GLOBAL_IMPLICIT_VALUE and INI_GLOBAL_IMPLICIT_V_LEN are used only as placeholders for custom information accessible solely by the user, starting from version 1.14.0 it is safer to make INI_GLOBAL_IMPLICIT_V_LEN match exactly the real length of INI_GLOBAL_IMPLICIT_VALUE (without counting the NUL terminator). By doing so it is possible to make libconfini aware of a segment of memory that must be protected from writing operations.
After having set the value to be assigned to implicit key elements and having enabled IniFormat::implicit_is_not_empty in the format, it is possible to test whether a dispatched key is implicit or not by comparing the address of its value property with the global variable INI_GLOBAL_IMPLICIT_VALUE:
Implicit keys can be parsed as booleans also without setting IniFormat::implicit_is_not_empty to true. By doing so there will be no distinction between empty and implicit keys, and there are situations where this can be a wanted behavior. The following example will parse both my_key and my_key = in the INI file as true:
The functions load_ini_file(), load_ini_path(), strip_ini_cache(), ini_array_foreach() and ini_array_split() require some listeners defined by the user. Such listeners must return an int value. When this is non-zero the caller function is interrupted, its loop stopped, and a non-zero value is returned by the caller as well.
The functions load_ini_file(), load_ini_path() and strip_ini_cache() return a non-zero value also if the INI file, for any reason, has not been completely parsed (see enum ConfiniInterruptNo). Therefore, in order to be able to distinguish between internal errors and user-generated interruptions the mask CONFINI_ERROR can be used.
For instance, in the following example the f_foreach() listener returns a non-zero value if a key named password with a value that equals Hello world is found. Hence, by using the mask CONFINI_ERROR, the code below distinguishes a non-zero value generated by the listener from a non-zero value due to a parsing error.
The functions ini_unquote(), ini_string_parse(), ini_array_collapse(), ini_array_break(), ini_array_release() and ini_array_split() change the content of the given strings. It is important to point out that the edit is always performed within the lengths of the strings given.
The behavior of these functions depends on the format used. In particular, using ini_string_parse() as a model one obtains the following scheme:
!format.no_single_quotes && !format.no_double_quotes && format.multiline_nodes != INI_NO_MULTILINE\\, \", \'ini_string_parse(): \\, \' and \" will be unescaped, all unescaped single and double quotes will be removed, then the new length of the string will be returned.!format.no_single_quotes && format.no_double_quotes && format.multiline_nodes != INI_NO_MULTILINE\\, \'ini_string_parse(): \\ and \' will be unescaped, all unescaped single quotes will be removed, then the new length of the string will be returned.format.no_single_quotes && !format.no_double_quotes && format.multiline_nodes != INI_NO_MULTILINE\\, \"ini_string_parse(): \\ and \" will be unescaped, all unescaped double quotes will be removed, then the new length of the string will be returned.format.no_single_quotes && format.no_double_quotes && format.multiline_nodes != INI_NO_MULTILINE\\ini_string_parse(): only \\ will be unescaped, spaces at the beginning and at the end of the string will be removed, then the new length of the string will be returned.format.no_single_quotes && format.no_double_quotes && format.multiline_nodes == INI_NO_MULTILINEini_string_parse(): Spaces at the beginning and at the end of the string will be removed, then the new length of the string will be returned.A function-like macro named INIFORMAT_HAS_NO_ESC() is available for checking whether a format supports escape sequences or not.
It is possible to extend the list of supported escape sequences by parsing additional ones before invoking ini_string_parse(). Under examples/utilities/ini_string_preparse.h there is a little helper function that adds support to the following sequences: "\a", "\b", "\f", "\n", "\r", "\t", "\v" and "\e". That function must be used right before ini_unquote() or ini_string_parse(), as it relies on the latter for unescaping double backslashes:
The native formatting functions ini_array_break(), ini_array_collapse(), ini_array_release(), ini_array_split(), ini_string_parse() and ini_unquote() have a special safeguard against attempting to edit the global variable INI_GLOBAL_IMPLICIT_VALUE. But if you use your own functions for editing the dispatches received, you must always make sure that dispatch->value != INI_GLOBAL_IMPLICIT_VALUE in formats where IniFormat::implicit_is_not_empty is set to true – otherwise a “Segmentation fault (core dumped)” error can be generated.
To check whether a particular substring belongs to INI_GLOBAL_IMPLICIT_VALUE, the INI_IS_IMPLICIT_SUBSTR() macro can be used:
In order to be as flexible as possible, libconfini does not store the dispatched data, nor indicizes them. This gives developers the power to deal with them in many different ways.
For small INI files a normal if/else chain, using ini_array_match() for comparing section paths and ini_string_match_si()/ini_string_match_ii() for comparing key names, usually represents the most practical and efficient way to obtain the information required from an INI file. Sometimes however, especially in case of sizeable INI files, the most efficient solution would be to store the parsed data in a hash table before trying to access it.
Some INI parsers are released with a hash table API included by default. This is an unpractical solution, since fantastic free software libraries that focus solely on hash tables already exist, and providing a further API for managing a hash function together with an INI parser only complicates the code, makes it harder to maintain, and does not give the developer the real freedom to choose what suits best to each single case. Some programming languages even have hash tables in their standard libraries (see std::map in C++ for example).
When needed, the data parsed by libconfini can be stored in a hash table while it is being dispatched. If you are interested in combining libconfini with a hash table, we have left a general example of how to use GLib's GHashTable together with libconfini under examples/miscellanea/glib_hash_table.c. If you are using C++, you can find an example of how to construct a C++ class that relies on a std::unordered_map object under examples/cplusplus/map.cpp. By keeping these examples as models other solutions can be easily explored as well.
Writing
will always allocate the smallest buffer large enough to store all the IniDispatch::data and IniDispatch::value received, including their NUL terminators and any possible implicit value.
This buffer will not be capable enough to store also the IniDispatch::append_to strings received (about which no information is available at early stages). However, thanks to the syntax of INI files, where a key is always appended to the section previously dispatched, it will be always possible to store the information about the structure of the tree by using numbers as references to the parents instead of copying verbatim the IniDispatch::append_to strings. This will also result in a more efficient code.
In the past, besides the two global variables INI_GLOBAL_IMPLICIT_VALUE and INI_GLOBAL_IMPLICIT_V_LEN, a third variable named INI_GLOBAL_LOWERCASE_MODE would tell libconfini whether to dispatch in lower case all key names and section paths of case-insensitive INI files. This variable rarely needed to be set to true, since string comparisons made by libconfini are always either case-sensitive or case-insensitive depending on the format given.
This “dispatch in lowercase” functionality itself is a relic from a time when libconfini did not yet possess any string comparison functions. This is no longer the case, and there are now four functions for that; therefore starting from version 1.15.0 both the INI_GLOBAL_IMPLICIT_VALUE variable and the ini_global_set_lowercase_mode() function have been marked as deprecated (see the list of deprecated functions and variables). If needed, to convert to lowercase key names and section paths in case-insensitive INI files it is possible to use the following simple snippet, which maps exactly what libconfini would do:
Comparing an ASCII upper case letter to an ASCII lower case letter is an invariant process. But comparing two Unicode letter cases is a process that depends on the locale of the machine. Consider for example the lower case letter i: in most European languages its upper case is I, while this is not the case in Turkish, where the upper case of i is İ (and the lower case of I is ı). Therefore for a person living in Italy or France, i and I will represent the same letter, while for a person living in Turkey they will not.
Key and section names of an INI file however cannot depend on the locale of the machine, since they must be reliably searched for independently of where a machine is located. Imagine for example a key named “INI” and imagine that Unicode case folding were performed on key names during string comparisons. If you lived in Europe you could look up for such key using its lower case “ini”, while if you lived in Turkey you would have to use the lower case “ını” to find it. So the only solution is to consider Unicode characters out of the ASCII range always as case-sensitive. For this reason, libconfini (and probably any judicious INI parser) will never perform case folding of Unicode characters out of the ASCII range within key and section names.
It must be said that most Unicode characters do not possess a lower and upper case, and most characters outside of the ASCII range could theoretically appear without problems in key and section names also in case-insensitive INI files (think of the character § for example). And, as for case-sensitive INI files, no Unicode character would ever represent a problem. Nonetheless, it is generally more acceptable to use ASCII only within key and section names – and possibly, if needed, non-ASCII Unicode characters within values and comments.
That said, libconfini deals perfectly fine with UTF-8 (but is always case-sensitive outside of the ASCII range), so use the latter as you feel appropriate.
Depending on the format of the INI file, libconfini may use up to three global variables (INI_GLOBAL_IMPLICIT_VALUE, INI_GLOBAL_IMPLICIT_V_LEN and INI_GLOBAL_LOWERCASE_MODE). In order to be thread-safe these three variables (if needed) must be defined only once (either directly, or by using their setter functions ini_global_set_implicit_value() and ini_global_set_lowercase_mode()), or otherwise a mutex logic must be introduced.
Apart from the three variables above, each parsing allocates and frees its own memory and every function is fully reentrant, therefore the library can be considered thread-safe.
The philosophy of libconfini is that of parsing as much as possible without generating error exceptions. No parsing errors are returned once an INI file has been correctly allocated into the heap, with the exception of the out-of-range error CONFINI_EOOR (see enum ConfiniInterruptNo), whose meaning is that the dispatches are for unknown reasons more than expected – this error is possibly generated by the presence of bugs in the library's code and should never be returned (please contact me if this happens).
When an INI node is wrongly written in respect to the format given, it is dispatched verbatim as an INI_UNKNOWN node – see enum IniNodeType. Empty lines, or lines containing only spaces and empty quotes (if the latter are supported) will be skipped.
In order to avoid error exceptions, strings containing an unterminated quote will always be treated as if they had a virtual quote as their last + 1 character. For example,
will always determine the same behavior as if it were
Any format containing the following three settings will never produce INI_UNKNOWN nodes, even if instead of an INI file we tried to parse a .jpeg image (or anything else):
The algorithms used by libconfini stand in a delicate equilibrium between flexibility, speed and code readability, with flexibility as primary target. Performance can vary with the format used to parse an INI file, but in most cases is not a concern.
One can measure the performance of the library by doing something like:
By changing the format of the INI file on the code above one might obtain different results. In particular, switching disabled entry recognition off – by setting IniFormat::semicolon_marker and IniFormat::hash_marker to INI_ONLY_COMMENT or INI_IGNORE – and making the format non-multi-line – by setting IniFormat::multiline_nodes to INI_NO_MULTILINE – will have a positive impact on the performance.
On my laptop libconfini seems to parse around 95 MiB per second using the model format INI_DEFAULT_FORMAT. Whether this is enough for you or not, it depends only on your needs.
If you are interested in testing yourself the library's performance on a particular hardware, you can find a performance test under dev/tests/performance.
It is hard to imagine a reason to be interested in disabled entries if not for writing a GUI editor for INI files. If this is the case and you are not using libconfini like normal people do, you might wonder how to ensure that disabled entries and comments be always parsed without ambiguity.
In most of the cases libconfini is smart enough to distinguish a disabled entry from a comment. However some INI files can be tricky and might require some workarounds. For instance, imagine the following INI file:
And imagine that for unknown reasons the author of the INI file wanted only ;foo = bar to be considered as a disabled entry, and the first and last line as normal comments.
If we tried to parse it according to the format used below
we would obtain the following result:
As you can see, all comments but now=Sunday April 3rd, 2016 would be parsed as disabled entries – which is not what the author intended. Therefore, to ensure that such INI file be parsed properly, you can follow two possible approaches.
1. Intervene on the INI file. The reason why now=Sunday April 3rd, 2016 has been properly parsed as a comment – despite it really looks like a disabled entry – is because it has been nested within a comment block opened by more than one leading marker (in this case the two ##). As a general rule, libconfini never parses a comment beginning with more than one leading marker as a disabled entry, therefore this is the surest way to ensure that proper comments are always considered as such.
Hence, by adding one more number sign to the first comment
we obtain the wanted result:
2. Intervene on the format. There are cases where the INI file is automatically generated by machines (comments included), or distributed as such, and human intervention would be required on each machine-generated release of the INI file. In these cases – and if we are sure about the expected content of the INI file – we can restrict the format chosen in order to parse comments and disabled entries properly. In particular, the following fields of the IniFormat bitfield can have an impact on the disambiguation between comments and disabled entries.
Reliable general patterns:
IniFormat::semicolon_marker and IniFormat::hash_marker – The imaginary author of our INI file, if we observe the latter closer, chose the semicolon symbol as the marker of disabled entries and the hash symbol as the marker of comments. We may exploit this difference and set our my_format.semicolon_marker to INI_DISABLED_OR_COMMENT and our my_format.hash_marker to INI_ONLY_COMMENT to obtain the correct disambiguation. If you believe that this solution is too artificial, think that /etc/samba/smb.conf and /etc/pulse/daemon.conf are systematically distributed using this pattern.IniFormat::disabled_after_space – Setting this property to false, due to the initial space that follows the comment marker (# INI...), forces the entire line to be considered as a comment. Some authors use this syntax to distinguish between comments and disabled entries (examples are /etc/pacman.conf and /etc/bluetooth/main.conf)Temporary workarounds:
IniFormat::no_spaces_in_names – If our INI file has only comments containing more than one word and we are sure that key and section names cannot contain internal white spaces, we can set this property to true to enhance disambiguation.IniFormat::disabled_can_be_implicit – This property, if set to false, forces all comments that do not contain a key-value delimiter never to be considered as disabled entries. Despite not having an impact on our example, it has an impact on the disambiguation algorithms used by libconfini. Its value in INI_DEFAULT_FORMAT is set to false.As a general rule, libconfini will always try to parse as a disabled entry whatever comment is allowed (by the format) to contain one. Only if this attempt fails, the block will be dispatched as a normal comment.
1.9.1