inweb-bootstrap/foundation-module/Chapter 7/Version Numbers.w

[VersionNumbers::] Version Numbers.

Semantic version numbers such as 3.7.1.

@h Standard adoption.
The Semantic Version Number standard, semver 2.0.0, provides a strict set
of rules for the format and meaning of version numbers: see |https://semver.org|.

Prior to the standard most version numbers in computing usage looked like
dot-divided runs of non-negative integers: for example, 4, 7.1, and 0.2.3.
The standard now requires exactly three: major, minor and patch. It's
therefore formally incorrect to have a version 2, or a version 2.3. We will
not be so strict on the textual form, which we will allow to be abbreviated.
Thus:

(a) The text |6.4.7| is understood to mean 6.4.7 and printed back as |6.4.7|
(b) The text |6| is understood to mean 6.0.0 and printed back as |6|
(c) The text |6.1| is understood to mean 6.1.0 and printed back as |6.1|
(d) The text |6.1.0| is understood to mean 6.1.0 and printed back as |6.1.0|

Similarly, the absence of a version number (called "null" below) will be
understood to mean 0.0.0, but will be distinguished from the explicit choice
to number something as 0.0.0.

@ A complication is that Inform 7 extensions have for many years allowed two
forms of version number: either just |N|, which fits the scheme above, or
|N/DDDDDD|, which does not. This is a format which was chosen for sentimental
reasons: IF enthusiasts know it well from the banner text of the Infocom
titles of the 1980s. This story file, for instance, was compiled at the
time of the Reykjavik summit between Presidents Gorbachev and Reagan:
= (text)
	Moonmist
	Infocom interactive fiction - a mystery story
	Copyright (c) 1986 by Infocom, Inc. All rights reserved.
	Moonmist is a trademark of Infocom, Inc.
	Release number 9 / Serial number 861022
=
Story file collectors customarily abbreviate this in catalogues to |9/861022|.

We will therefore allow this notation, and convert it silently each way.
|N/DDDDDD| is equivalent to |N.DDDDDD|. Thus, |9/861022| means 9.861022.0 in
semver precedence order.

In all non-textual respects, and in particular on precedence rules, we follow
the standard exactly. The only reason we allow these abbreviations is because
we don't want to force Inform extension writers to type "Version 3.4.1 of
Such-and-Such by Me begins here", and so on: it would break all existing
extensions, for one thing, and it looks unfriendly.

@ In the array below, unspecified numbers are stored as |-1|. The three
components are otherwise required to be non-negative integers.

Semver allows for more elaborate forms: for example |3.1.41-alpha.72.zeta+6Q45|
would mean 3.1.41 but with prerelease versioning |alpha.72.zeta| and build
metadata |6Q45|. The |prerelease_segments| list for this would be a list of
three texts: |alpha|, |72|, |zeta|.

@d SEMVER_NUMBER_DEPTH 3 /* major, minor, patch */

=
typedef struct semantic_version_number {
	int version_numbers[SEMVER_NUMBER_DEPTH];
	struct linked_list *prerelease_segments; /* of |text_stream| */
	struct text_stream *build_metadata;
} semantic_version_number;

typedef struct semantic_version_number_holder {
	struct semantic_version_number version;
	CLASS_DEFINITION
} semantic_version_number_holder;

@ All invalid strings of numbers -- i.e., breaking the above rules -- are
called "null" versions, and can never be valid as the version of anything.
Instead they are used to represent the absence of a version number.
(In particular, a string of |-1|s is null.)

=
semantic_version_number VersionNumbers::null(void) {
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wconditional-uninitialized"
	semantic_version_number V;
	for (int i=0; i<SEMVER_NUMBER_DEPTH; i++) V.version_numbers[i] = -1;
	V.prerelease_segments = NULL;
	V.build_metadata = NULL;
	return V;
#pragma clang diagnostic pop
}

int VersionNumbers::is_null(semantic_version_number V) {
	for (int i=0, allow=TRUE; i<SEMVER_NUMBER_DEPTH; i++) {
		if (V.version_numbers[i] < -1) return TRUE; /* should never happen */
		if (V.version_numbers[i] == -1) allow = FALSE;
		else if (allow == FALSE) return TRUE; /* should never happen */
	}
	if (V.version_numbers[0] < 0) return TRUE;
	return FALSE;
}

@h Printing and parsing.
Printing is simple enough:

=
void VersionNumbers::to_text(OUTPUT_STREAM, semantic_version_number V) {
	if (VersionNumbers::is_null(V)) { WRITE("null"); return; }
	for (int i=0; (i<SEMVER_NUMBER_DEPTH) && (V.version_numbers[i] >= 0); i++) {
		if (i>0) WRITE(".");
		WRITE("%d", V.version_numbers[i]);
	}
	if (V.prerelease_segments) {
		int c = 0;
		text_stream *T;
		LOOP_OVER_LINKED_LIST(T, text_stream, V.prerelease_segments) {
			if (c++ == 0) WRITE("-"); else WRITE(".");
			WRITE("%S", T);
		}			
	}
	if (V.build_metadata) WRITE("+%S", V.build_metadata);
}

@ And this provides for the |%v| escape, though we must be careful when
using this to pass a pointer to the version, not the version itself;
variadic macros are not carefully enough type-checked by |clang| or |gcc|
to catch this sort of slip.

=
void VersionNumbers::writer(OUTPUT_STREAM, char *format_string, void *vE) {
	semantic_version_number *V = (semantic_version_number *) vE;
	VersionNumbers::to_text(OUT, *V);
}

@ Parsing is much more of a slog. The following returns a null version if
the text |T| is in any respect malformed, i.e., if it deviates from the
above specification in even the most trivial way. We parse the three parts
of a semver version in order: e.g. |3.1.41-alpha.72.zeta+6Q45| the first
part is up to the hyphen, the second part between the hyphen and the plus
sign, and the third part runs to the end. The second and third parts are
optional, but if both are given, they must be in that order.

@e MMP_SEMVERPART from 1
@e PRE_SEMVERPART
@e BM_SEMVERPART

=
semantic_version_number VersionNumbers::from_text(text_stream *T) {
	semantic_version_number V = VersionNumbers::null();
	int component = 0, val = -1, dots_used = 0, slashes_used = 0, count = 0;
	int part = MMP_SEMVERPART;
	TEMPORARY_TEXT(prerelease)
	LOOP_THROUGH_TEXT(pos, T) {
		wchar_t c = Str::get(pos);
		switch (part) {
			case MMP_SEMVERPART: 
				if (c == '.') dots_used++;
				if (c == '/') slashes_used++;
				if ((c == '.') || (c == '/') || (c == '-') || (c == '+')) {
					if (val == -1) return VersionNumbers::null();
					if (component >= SEMVER_NUMBER_DEPTH) return VersionNumbers::null();
					V.version_numbers[component] = val;
					component++; val = -1; count = 0;
					if (c == '-') part = PRE_SEMVERPART;
					if (c == '+') part = BM_SEMVERPART;
				} else if (Characters::isdigit(c)) {
					int digit = c - '0';
					if ((val == 0) && (slashes_used == 0))
						return VersionNumbers::null();
					if (val < 0) val = digit; else val = 10*val + digit;
					count++;
				} else return VersionNumbers::null();
				break;
			case PRE_SEMVERPART:
				if (c == '.') {
					@<Add prerelease content@>;
				} else if (c == '+') {
					@<Add prerelease content@>; part = BM_SEMVERPART;
				} else {
					PUT_TO(prerelease, c);
				}
				break;
			case BM_SEMVERPART:
				if (V.build_metadata == NULL) V.build_metadata = Str::new();
				PUT_TO(V.build_metadata, c);
				break;
		}
	}
	if ((part == PRE_SEMVERPART) && (Str::len(prerelease) > 0)) @<Add prerelease content@>;
	DISCARD_TEXT(prerelease)
	if ((dots_used > 0) && (slashes_used > 0)) return VersionNumbers::null();
	if (slashes_used > 0) {
		if (component > 1) return VersionNumbers::null();
		if (count != 6) return VersionNumbers::null();
	}
	if (part == MMP_SEMVERPART) {
		if (val == -1) return VersionNumbers::null();
		if (component >= SEMVER_NUMBER_DEPTH) return VersionNumbers::null();
		V.version_numbers[component] = val;
	}
	return V;
}

@<Add prerelease content@> =
	if (Str::len(prerelease) == 0) return VersionNumbers::null();
	if (V.prerelease_segments == NULL) V.prerelease_segments = NEW_LINKED_LIST(text_stream);
	ADD_TO_LINKED_LIST(Str::duplicate(prerelease), text_stream, V.prerelease_segments);
	Str::clear(prerelease);

@h Precendence.
The most important part of the semver standard is the rule on which versions
take precedence over which others, and we follow it exactly. The following
criteria are used in turn: major version; minor version; patch version;
any prerelease elements, which must be compared numerically if consisting
of digits only, and alphabetically otherwise; and finally the number of
prerelease elements. Build metadata is disregarded entirely.

=
int VersionNumbers::le(semantic_version_number V1, semantic_version_number V2) {
	for (int i=0; i<SEMVER_NUMBER_DEPTH; i++) {
		int N1 = VersionNumbers::floor(V1.version_numbers[i]);
		int N2 = VersionNumbers::floor(V2.version_numbers[i]);
		if (N1 > N2) return FALSE;
		if (N1 < N2) return TRUE;
	}
	linked_list_item *I1 = (V1.prerelease_segments)?(LinkedLists::first(V1.prerelease_segments)):NULL;
	linked_list_item *I2 = (V2.prerelease_segments)?(LinkedLists::first(V2.prerelease_segments)):NULL;
	while ((I1) && (I2)) {
		text_stream *T1 = (text_stream *) LinkedLists::content(I1);
		text_stream *T2 = (text_stream *) LinkedLists::content(I2);
		int N1 = VersionNumbers::strict_atoi(T1);
		int N2 = VersionNumbers::strict_atoi(T2);
		if ((N1 >= 0) && (N2 >= 0)) {
			if (N1 < N2) return TRUE;
			if (N1 > N2) return FALSE;
		} else {
			if (Str::ne(T1, T2)) {
				int c = Str::cmp(T1, T2);
				if (c < 0) return TRUE;
				if (c > 0) return FALSE;
			}
		}
		I1 = LinkedLists::next(I1);
		I2 = LinkedLists::next(I2);
	}
	if ((I1 == NULL) && (I2)) return TRUE;
	if ((I1) && (I2 == NULL)) return FALSE;
	return TRUE;
}

@ The effect of this is to read unspecified versions of major, minor or patch
as if they were 0:

=
int VersionNumbers::floor(int N) {
	if (N < 0) return 0;
	return N;
}

@ This returns a non-negative integer if |T| contains only digits, and |-1|
otherwise. If the value has more than about 10 digits, then the result will
not be meaningful, which I think is a technical violation of the standard.

=
int VersionNumbers::strict_atoi(text_stream *T) {
	LOOP_THROUGH_TEXT(pos, T)
		if (Characters::isdigit(Str::get(pos)) == FALSE)
			return -1;
	wchar_t c = Str::get_first_char(T);
	if ((c == '0') && (Str::len(T) > 1)) return -1;
	return Str::atoi(T, 0);
}

@h Trichotomy.
We now use the above function to construct ordering relations on semvers.
These are trichotomous, that is, for each pair |V1, V2|, exactly one of the
|VersionNumbers::eq(V1, V2)|, |VersionNumbers::gt(V1, V2)|, |VersionNumbers::lt(V1, V2)|
is true.

=
int VersionNumbers::eq(semantic_version_number V1, semantic_version_number V2) {
	if ((VersionNumbers::le(V1, V2)) && (VersionNumbers::le(V2, V1)))
		return TRUE;
	return FALSE;
}

int VersionNumbers::ne(semantic_version_number V1, semantic_version_number V2) {
	return (VersionNumbers::eq(V1, V2))?FALSE:TRUE;
}

int VersionNumbers::gt(semantic_version_number V1, semantic_version_number V2) {
	return (VersionNumbers::le(V1, V2))?FALSE:TRUE;
}

int VersionNumbers::ge(semantic_version_number V1, semantic_version_number V2) {
	return VersionNumbers::le(V2, V1);
}

int VersionNumbers::lt(semantic_version_number V1, semantic_version_number V2) {
	return (VersionNumbers::ge(V1, V2))?FALSE:TRUE;
}

@ And the following can be used for sorting, following the |strcmp| convention.

=
int VersionNumbers::cmp(semantic_version_number V1, semantic_version_number V2) {
	if (VersionNumbers::eq(V1, V2)) return 0;
	if (VersionNumbers::gt(V1, V2)) return 1;
	return -1;
}