diff options
Diffstat (limited to 'doc/context/sources/general/manuals/lowlevel/lowlevel-expansion.tex')
| -rw-r--r-- | doc/context/sources/general/manuals/lowlevel/lowlevel-expansion.tex | 290 |
1 files changed, 241 insertions, 49 deletions
diff --git a/doc/context/sources/general/manuals/lowlevel/lowlevel-expansion.tex b/doc/context/sources/general/manuals/lowlevel/lowlevel-expansion.tex index 433632e2d..98f533859 100644 --- a/doc/context/sources/general/manuals/lowlevel/lowlevel-expansion.tex +++ b/doc/context/sources/general/manuals/lowlevel/lowlevel-expansion.tex @@ -1,5 +1,8 @@ % language=us runpath=texruns:manuals/lowlevel +% This is work in progress and after an initial draft got extended because of the +% 2021 meeting. It will hopefully improve over time. + \usemodule[system-tokens] \environment lowlevel-style @@ -10,13 +13,15 @@ \startsection[title=Preamble] -% \startsubsection[title=Introduction] -% \stopsubsection - This short manual demonstrates a couple of properties of the macro language. It -is not the in|-|depth philosophical expose about macro languages, tokens, +is not an in|-|depth philosophical expose about macro languages, tokens, expansion and such that some \TEX ies like. I prefer to stick to the practical -aspects. +aspects. Occasionally it will be technical but you can just skip those paragraphs +(or later return to them) when you can't follow the explanation. It's often not +that relevant. I won't talk in terms of mouth, stomach and gut the way the \TEX +book does and although there is no way to avoid the word \quote {token} I will do +my best to not complicate matters by too much token speak. Examples show best +what we mean. \stopsection @@ -24,8 +29,8 @@ aspects. The \TEX\ language provides quite some commands and those built in are called primitives. User defined commands are called macros. A macro is a shortcut to a -list of primitives, macro calls. All can be mixed with characters that are to -be typeset somehow. +list of primitives and|/|or macro calls. All can be mixed with characters that +are to be typeset somehow. \starttyping[option=TEX] \def\MyMacro{b} @@ -40,9 +45,79 @@ macro. In this case it sees just an \type {b} and it will give this the same treatment as the \type {a}. The macro ends, the input level decrements and the \type {c} gets its treatment. -A macro can contain references to macros so in practice the input can go several -levels up and some applications push back a lot so this is why your \TEX\ input -stack can be configured. +Before we move on to more examples and differences between engines, it is good to +stress that \type {\MyMacro} is not a primitive command: we made our command +here. The \type {b} actually can be seen as a sort of primitive because in this +macro it gets stored as so called token with a primitive property. That primitive +property can later on be used to determine what to do. More explicit examples of +primitives are \type {\hbox}, \type {\advance} and \type {\relax}. It will be +clear that \CONTEXT\ extends the repertoire of primitive commands with a lot of +macro commands. When we typeset a source using module \type {m-scite} the +primitives come out dark blue. + +The amount of primitives differs per engine. It all starts with \TEX\ as written +by Don Knuth. Later \ETEX\ added some more primitives and these became official +extensions adopted by other variants of \TEX. The \PDFTEX\ engine added quite +some and as follow up on that \LUATEX\ added more but didn't add all of \PDFTEX. +A few new primitives came from \OMEGA\ (\ALEPH). The \LUAMETATEX\ engine drops a +set of primitives that comes with \LUATEX\ and adds plenty new ones. The nature +of this engine (no backend and less frontend) makes that we need to implement +some primitives as macros. But the basic set is what good old \TEX\ comes with. + +Internally these so called primitives are grouped in categories that relate to +their nature. They can be directly expanded (a way of saying that they get +immediately interpreted) or delayed (maybe stored for later usage). They can +involve definitions, calculations, setting properties and values or they can +result in some typesetting. This is what makes \TEX\ confusing to new users: it +is a macro programming language, an interpreter but at the same time an executor +of typesetting instructions. + +A group of primitives is internally identified as a command (they have a \type +{cmd} code) and the sub commands are flagged by their \type {chr} code. This +sounds confusing but just thing of the fact that most of what we input are +characters and therefore they make up most sub commands. For instance the \quote +{letter \type {cmd}} is used for characters that are seen as letters that can be +used in the name of user commands, can be typeset, are valid for hyphenation +etc.\ The letter related \type {cmd} can have many \type {chr} codes (all of +\UNICODE). I'd like to remark that the grouping is to a large extend functional, +so sometimes primitives that you expect to be similar in nature are in different +groups. This has to do with the fact that \TEX\ needs to be a able to determine +efficiently if a primitive is operating (or forbidden) in horizontal, vertical +and|/|or math mode. + +There are more than 150 internal \type {cmd} groups. if we forget about the +mentioned character related ones, some, have only a few sub commands (\type +{chr}) and others many more (just consider all the \OPENTYPE\ math spacing +related parameters). A handful of these commands deal with what we call macros: +user defined combinations of primitives and other macros, consider them little +programs. The \type {\MyMacro} example above is an example. There are differences +between engines. In standard \TEX\ there are \type {\outer} and \type {\long} +commands, and most engines have these. However, in \LUAMETATEX\ the later to be +discussed \type {\protected} macros have their own specific \quote {call \type +{cmd}}. Users don't need to bother about this. + +So, when from now on we talk about primitives, we mean the built in, hard coded +commands, and when we talk about macros we mean user commands. Although +internally there are less \type {cmd} categories than primitives, from the +perspective of the user they are all unique. Users won't consult the source +anyway but when they do they are warned. Also, when in \LUAMETATEX\ you use the +low level interfacing to \TEX\ you have to figure out these subtle aspects +because there this grouping does matter. + +Before we continue I want to make clear that expansion (as discussed in this +document) can refer to a macro being expanded (read: its meaning gets injected +into the input, so the engine kind of sidetracks from what is was doing) but also +to direct consequences of running into a primitive. However, users only need to +consider expansion in the perspective of macros. If a user has \type {\advance} +in the input it immediately gets done. But when it's part of a macro definition +it only is executed when the macro expands. A good check in (traditional) \TEX\ +is to compare what happens in \type {\def} and \type {\edef} which is why we will +use these two in the upcoming examples. You put something in a macro and then +check what \type {\meaning} or \type {\show} reports. + +Now back to user defined macros. A macro can contain references to macros so in +practice the input can go several levels up and some applications push back a lot +so this is why your \TEX\ input stack can be configured to be huge. \starttyping[option=TEX] \def\MyMacroA{ and } @@ -52,11 +127,11 @@ a\MyMacroA b \stoptyping When \type {\MyMacroB} is defined, its body gets three so called tokens: the -character token \type {a} with property \quote {other}, a token that is a +character token \type {1} with property \quote {other}, a token that is a reference to the macro \type {\MyMacroB}, and a character token \type {2}, also -with property \quote {other} The meaning of \type {\MyMacroA} became five tokens: +with property \quote {other} The meaning of \type {\MyMacroA} is five tokens: a reference to a space token, then three character tokens with property \quote -{letter}, and finally again a space token. +{letter}, and finally a space token. \starttyping[option=TEX] \def \MyMacroA{ and } @@ -65,8 +140,8 @@ a reference to a space token, then three character tokens with property \quote a\MyMacroA b \stoptyping -In the previous example an \type {\edef} is used, where the \type {e} indicates -expansion. This time the meaning gets expanded. So we get effectively the same +In the second definition an \type {\edef} is used, where the \type {e} indicates +expansion. This time the meaning gets expanded immediately. So we get effectively the same as in: \starttyping[option=TEX] @@ -123,7 +198,7 @@ Let's assume that \type {\scratchcounter} is zero to start with and use \type \meaning\MyMacroC \stoplines -So, this time the third macro has basically its meaning frozen, but we can +So, this time the third macro has its meaning frozen, but we can prevent this by applying a \type {\noexpand} when we do this: \startbuffer @@ -156,7 +231,7 @@ their meanings). \def\MyShow{\quotation {\strut \inlinebuffer \expandafter \typ \expandafter {\the\scratchtoks}\strut}} -You can also store tokens in a so called token register. Here we use a predefined +You can also store tokens in a so-called token register. Here we use a predefined scratch register: \startbuffer @@ -199,12 +274,12 @@ Indeed the macro gets expanded but only one level: \MyShow. Compare this with: \typebuffer[option=TEX] -The trick is to expand in two steps: \MyShow. Later we will see that other +The trick is to expand in two steps with an intermediate \type {\edef}: \MyShow. Later we will see that other engines provide some more expansion tricks. The only way to get some grip on expansion is to just play with it. -The \type {\expandafter} primitive expands the token (which can be a macro) after -the next next one and injects its meaning into the stream. So: +The \type {\expandafter} primitive expands the token (which can be a macro) standing after +the next next one and then injects its meaning into the stream. So: \starttyping[option=TEX] \expandafter \MyMacroA \MyMacroB @@ -271,7 +346,7 @@ These macros have the following meanings: \meaning\MyMacroABC \stoplines -In \CONTEXT\ you will use the \type {\unexpanded} prefix instead because that one +In \CONTEXT\ you will use the \type {\unexpanded} prefix instead, because that one did something similar in older versions of \CONTEXT. As we were early adopters of \ETEX, this later became a synonym to the \ETEX\ primitive. @@ -286,17 +361,17 @@ did something similar in older versions of \CONTEXT. As we were early adopters o Here the wrapper around the token register assignment will expand the three macros, unless they are protected, so its content becomes \MyShow. This saves -either a lot of more complex \type {\expandafter} usage or using an intermediate +either a lot of more complex \type {\expandafter} usage or the need to use an intermediate \type {\edef}. In \CONTEXT\ the \type {\expanded} macro does something simpler -but it doesn't expand the first token as it is meant as a wrapper around a command, +but it doesn't expand the first token as this is meant as a wrapper around a command, like: \starttyping[option=TEX] \expanded{\chapter{....}} % a ConTeXt command \stoptyping -where we do want to expand the title but not the \type {\chapter} command, not -that this would happen actually because \type {\chapter} is a protected command. +where we do want to expand the title but not the \type {\chapter} command (not +that this would happen actually because \type {\chapter} is a protected command.) The counterpart of \type {\normalexpanded} is \type {\normalunexpanded}, as in: @@ -325,7 +400,7 @@ close, with: \typebuffer[option=TEX] This (when typeset monospaced) is: {\tt \inlinebuffer}. The detokenizer is like -\type {\string} applied to each token in its argument. Compare this: +\type {\string} applied to each token in its argument. Compare this to: \startbuffer \normalexpanded { @@ -447,8 +522,8 @@ We now get: \type{\TestD} : {\tttf \meaningless\TestD} \stoplines -There are however cases where one wishes this to happen automatically but that -will also make protected macros expand that create havoc, like switching fonts. +There are however cases where one wishes this to happen automatically, but that +will also make protected macros expand which will create havoc, like switching fonts. \startbuffer \def\TestA{A} @@ -473,8 +548,8 @@ This time \type {\TestC} looses its protection: \stoplines Actually adding \type {\fullyexpanded} would be trivial but it makes not much -sense to add that overhead (at least not now). This feature is experimental -anyway so it might go away when I see no real advantage. +sense to add the overhead (at least not now). This feature is experimental +anyway so it might go away when I see no real advantage from it. When you store something in a macro or token register you always need to keep an eye on category codes. A dollar in the input is normally treated as math shift, a @@ -494,19 +569,19 @@ but a character flagged as \quote {other}. A whole new category of commands has to do with so called local control. When \TEX\ scans and interprets the input, a process takes place that is called tokenizing: (sequences of) characters get a symbolic representation and travel -through the system as tokens. Often they immediately get interpreted and then -discarded, but when for instance you define a macro they end up as a linked list +through the system as tokens. Often they immediately get interpreted and are then +discarded. But when for instance you define a macro they end up as a linked list of tokens in the macro body. We already saw that expansion plays a role. In most -cases, unless \TEX\ is collecting tokens, the main action is dealt with in the so -called main loop. Something gets picked up from the input but can also be pushed -back, for instance because of some lookahead that didn't result in some action. -Quite some time is spent in pushing and popping from the so called input stack. +cases, unless \TEX\ is collecting tokens, the main action is dealt with in the so-called +main loop. Something gets picked up from the input but can also be pushed +back, for instance because of some lookahead that didn't result in an action. +Quite some time is spent in pushing and popping from the so-called input stack. When we are in \LUA, we can pipe back into the engine but all is collected till we're back in \TEX\ where the collected result is pushed into the input. Because \TEX\ is a mix of programming and action there basically is only that main loop. There is no real way to start a sub run in \LUA\ and do all kind of things -independent of the current run. This makes sense when you consider the mix: it +independent of the current one. This makes sense when you consider the mix: it would get too confusing. However, in \LUATEX\ and even better in \LUAMETATEX, we can enter a sort of local @@ -662,7 +737,7 @@ When you want to add some grouping, it quickly can become less pretty: \getbuffer[def]\getbuffer[use] -A single token alternative is available too and its usage us like this: +A single token alternative is available too and its usage is like this: \startbuffer \def\TestA{\scratchcounter=100 } @@ -752,7 +827,7 @@ context.runstring([[\setbox0\hbox{(Here \bf 1.2345)}]]) context.runstring([[\setbox0\hbox{(Here \bf %.3f)}]],1.2345) \stoptyping -Before we had the string runner this was the way to do it when staying in \LUA\ +Before we had \type {runstring} this was the way to do it when staying in \LUA\ was needed: \startbuffer[run] @@ -780,7 +855,7 @@ context("[8: %p]",tex.box[0].width) \start \getbuffer[run] \stop The order of flushing matters because as soon as something is not stored in a -token list or macro body, \TEX\ will typeset it. And as said, a lot this relates +token list or macro body, \TEX\ will typeset it. And as said, a lot of this relates to pushing stuff into the input which is stacked. Compare: \startbuffer[run] @@ -809,7 +884,7 @@ tex.pushlocal() context("[HERE 2]") tex.poplocal() You can expand a macro at the \LUA\ end with \type {token.expandmacro} which has a peculiar interface. The first argument has to be a string (the name of a macro) -or a user data (a valid macro token). This macro can be fed with parameters by +or a userdata (a valid macro token). This macro can be fed with parameters by passing more arguments: \starttabulate[|||] @@ -840,15 +915,15 @@ primitive(s): \typebuffer[run][option=TEX] -The \type {\tokenized} takes an optionally keyword and the examples above give: {\tttf +The \type {\tokenized} takes an optional keyword and the examples above give: {\tttf \getbuffer[run]}. The \LUATEX\ primitive \type {\scantextokens} which is a -variant of \ETEX's \type {\scantokens} operate under the current catcode regime +variant of \ETEX's \type {\scantokens} operates under the current catcode regime (the last one honors \type {\everyeof}). The difference with \type {\tokenized} is that this one first serializes the given token list (just like \type {\detokenize}). \footnote {The \type {\scan*tokens} primitives now share the same helpers as \LUA, but they should behave the same as in \LUATEX.} -With \type {\retokenized} the catcode table index is mandate (it saves a bit of +With \type {\retokenized} the catcode table index is mandatory (it saves a bit of scanning and is easier on intermixed \type {\expandafter} usage. There often are several ways to accomplish the same: @@ -865,10 +940,10 @@ often are several ways to accomplish the same: \getbuffer[run] Here the numbers show the relative performance of these methods. The \type -{\detokenize} and \type {\meaningless} win because they already know that that a +{\detokenize} and \type {\meaningless} win because they already know that a verbose serialization is needed. The last two first serialize and then reinterpret the resulting token list using the given catcode regime. The last one -is slowest because has to scan the keyword. +is slowest because it has to scan the keyword. There is however a pitfall here: @@ -884,7 +959,7 @@ There is however a pitfall here: \typebuffer[run][option=TEX] The outcome is different now because we have an expandable embedded macro call. -The fact that we expand in the last two primitives is also a reason why they are +The fact that we expand in the last two primitives is also the reason why they are \quote {slower}. \getbuffer[run] @@ -901,7 +976,7 @@ introduced in this section: \retokenized \notcatcodes{\MyTitle}\crlf \retokenized \notcatcodes{\semiexpanded{\MyTitle}}\crlf \tokenized catcodetable \notcatcodes{\MyTitle}\crlf -\tokenized catcodetable \notcatcodes{\semiexpanded{\MyTitle}}\crlf +\tokenized catcodetable \notcatcodes{\semiexpanded{\MyTitle}} \stopbuffer \typebuffer[run][option=TEX] @@ -911,7 +986,7 @@ This time compare the last four lines: \getbuffer[run] Of course the question remains to what extend we need this and eventually will -apply it in \CONTEXT. The \type {\detokenize} is used already. History shows that +apply in \CONTEXT. The \type {\detokenize} is used already. History shows that eventually there is a use for everything and given the way \LUAMETATEX\ is structured it was not that hard to provide the alternatives without sacrificing performance or bloating the source. @@ -926,6 +1001,123 @@ performance or bloating the source. \stopsection +\startsection[title=Dirty tricks] + +When I was updating this manual Hans vd Meer and I had some discussions about +expansion and tokenization related issues when combining of \XML\ processing with +\TEX\ macros where he did some manipulations in \LUA. In these mixed cases you +can run into catcode related problems because in \XML\ you want for instance a +\type {#} to be a hash mark (other character) and not an parameter identifier. +Normally this is handled well in \CONTEXT\ but of course there are complex cases +where you need to adapt. + +Say that you want to compare two strings (officially we should say token lists) +with mixed catcodes. Let's also assume that you want to use the normal \type +{\if} construct (which was part of the discussion). We start with defining +a test set. The reason that we present this example here is that we use +commands discussed in previous sections: + +\startbuffer[run] + \def\abc{abc} +\semiprotected \def\xyz{xyz} + \edef\pqr{\expandtoken\notcatcodes`p% + \expandtoken\notcatcodes`q% + \expandtoken\notcatcodes`r} + +1: \ifcondition\similartokens{abc} {def}YES\else NOP\fi (NOP) \quad +2: \ifcondition\similartokens{abc}{\abc}YES\else NOP\fi (YES) + +3: \ifcondition\similartokens{xyz} {pqr}YES\else NOP\fi (NOP) \quad +4: \ifcondition\similartokens{xyz}{\xyz}YES\else NOP\fi (YES) + +5: \ifcondition\similartokens{pqr} {pqr}YES\else NOP\fi (YES) \quad +6: \ifcondition\similartokens{pqr}{\pqr}YES\else NOP\fi (YES) +\stopbuffer + +\typebuffer[run][option=TEX] + +So, we have a mix of expandable and semi expandable macros, and also a mix of +catcodes. A naive approach would be: + +\startbuffer[def] +\permanent\protected\def\similartokens#1#2% + {\iftok{#1}{#2}} +\stopbuffer + +\typebuffer[def][option=TEX] + +but that will fail on some cases: + +\pushoverloadmode \startpacked \tttf \getbuffer[def,run]\stoppacked \popoverloadmode + +So how about: + +\startbuffer[def] +\permanent\protected\def\similartokens#1#2% + {\iftok{\detokenize{#1}}{\detokenize{#2}}} +\stopbuffer + +\typebuffer[def][option=TEX] + +That one is even worse: + +\pushoverloadmode \startpacked \tttf \getbuffer[def,run]\stoppacked \popoverloadmode + +We need to expand so we end up with this: + +\startbuffer[def] +\permanent\protected\def\similartokens#1#2% + {\normalexpanded{\noexpand\iftok + {\noexpand\detokenize{#1}} + {\noexpand\detokenize{#2}}}} +\stopbuffer + +\typebuffer[def][option=TEX] + +Better: + +\pushoverloadmode \startpacked \tttf \getbuffer[def,run]\stoppacked \popoverloadmode + +But that will still not deal with the mildly protected macro so in the end we +have: + +\startbuffer[def] +\permanent\protected\def\similartokens#1#2% + {\semiexpanded{\noexpand\iftok + {\noexpand\detokenize{#1}} + {\noexpand\detokenize{#2}}}} +\stopbuffer + +\typebuffer[def][option=TEX] + +Now we're good: + +\pushoverloadmode \startpacked \tttf \getbuffer[def,run]\stoppacked \popoverloadmode + +Finally we wrap this one in the usual \type {\doifelse...} macro: + +\startbuffer[def] +\permanent\protected\def\doifelsesimilartokens#1#2% + {\ifcondition\similartokens{#1}{#2}% + \expandafter\firstoftwoarguments + \else + \expandafter\secondoftwoarguments + \fi} +\stopbuffer + +\typebuffer[def][option=TEX] + +so that we can do: + +\starttyping[option=TEX] +\doifelsesimilartokens{pqr}{\pqr}{YES}{NOP} +\stoptyping + +A companion macro of this is \type {\wipetoken} but for that one you need to look +into the source. + +\stopsection + \stopdocument % \aftergroups |