Difference between revisions of "VM for transfer"

From Apertium
Jump to navigation Jump to search
(Link to French page)
 
(13 intermediate revisions by 5 users not shown)
Line 1: Line 1:
  +
[[Machine virtuelle pour le transfert|En français]]
  +
 
== Instruction Set ==
 
== Instruction Set ==
   
Line 16: Line 18:
 
| clip || - || N/A || part → value || Obtains the part in the only language there is (inter/post-chunk) and pushes the value onto the stack
 
| clip || - || N/A || part → value || Obtains the part in the only language there is (inter/post-chunk) and pushes the value onto the stack
 
|-
 
|-
| clipsl || - || N/A || part, pos → value || Obtains the 'part' in source language in position 'pos' and pushes the 'value' onto the stack
+
| clipsl || - || link-to || part, pos &rarr; value || Obtains the 'part' in source language in position 'pos' and pushes the 'value' onto the stack. An optional operand is used for clips with link-to tags, e.g. "clipsl <3>".
 
|-
 
|-
| cliptl || - || N/A || part, pos &rarr; value || Obtains the 'part' in target language in position 'pos' and pushes the 'value' onto the stack
+
| cliptl || - || link-to || part, pos &rarr; value || Obtains the 'part' in target language in position 'pos' and pushes the 'value' onto the stack. An optional operand is used for clips with link-to tags, e.g. "cliptl <3>".
 
|-
 
|-
 
| storecl || - || N/A || value, part &rarr; [empty] || Stores 'value' in the only language there is (inter/post-chunk)
 
| storecl || - || N/A || value, part &rarr; [empty] || Stores 'value' in the only language there is (inter/post-chunk)
Line 36: Line 38:
 
| lu-count || - || N/A || [empty] &rarr; number || Pushes the number of lexical units (words inside the chunk) in the rule onto the stack
 
| lu-count || - || N/A || [empty] &rarr; number || Pushes the number of lexical units (words inside the chunk) in the rule onto the stack
 
|-
 
|-
| chunk || - || N || elem<sub>N-1</sub>, ... , elem<sub>1</sub>, <tags>, name &rarr; ^name<tags>{elem<sub>1</sub>...elem<sub>N-1</sub>}$ || Pops 'N' amount of data from the stack, creates the chunk and pushes it back onto the stack
+
| chunk || - || N || elem<sub>N-2</sub>, ... , elem<sub>1</sub>, <tags>, name &rarr; ^name<tags>{elem<sub>1</sub>...elem<sub>N-2</sub>}$ || Pops 'N' amount of data from the stack, creates the chunk and pushes it back onto the stack
 
|-
 
|-
 
| out || - || N || value<sub>N</sub>, ..., value<sub>1</sub> &rarr; [empty] || Pops 'N' values from the stack and outputs them
 
| out || - || N || value<sub>N</sub>, ..., value<sub>1</sub> &rarr; [empty] || Pops 'N' values from the stack and outputs them
Line 88: Line 90:
 
* The case is represented as "aa" (all lowercase), "Aa" (first uppercase) and "AA", (all uppercase).
 
* The case is represented as "aa" (all lowercase), "Aa" (first uppercase) and "AA", (all uppercase).
   
  +
== Code generation ==
== Sample compilation of XML code fragments ==
 
   
=== Example 1 ===
+
=== Code sections ===
==== XML t1x Code: chunking ====
 
<code>
 
<out>
 
<chunk name="det_det_nom_adj" case="caseFirstWord">
 
<tags>
 
<tag><lit-tag v="SN"/></tag>
 
<tag><var n="tipus_det"/></tag>
 
<tag><var n="gen_chunk"/></tag>
 
<tag><var n="nbr_chunk"/></tag>
 
</tags>
 
<lu>
 
<clip pos="1" side="tl" part="lem"/>
 
<clip pos="1" side="tl" part="a_det"/>
 
<clip pos="1" side="tl" part="gen_sense_mf" link-to="3"/>
 
<clip pos="1" side="tl" part="gen_mf"/>
 
<clip pos="1" side="tl" part="nbr_sense_sp" link-to="4"/>
 
<clip pos="1" side="tl" part="nbr_sp"/>
 
</lu>
 
<b/>
 
<lu>
 
<lit v="el"/>
 
<lit-tag v="det.def"/>
 
<clip pos="1" side="tl" part="gen_sense_mf" link-to="3"/>
 
<lit-tag v="pl"/>
 
</lu>
 
<b pos="1"/>
 
<lu>
 
<clip pos="3" side="tl" part="lemh"/>
 
<clip pos="3" side="tl" part="a_nom"/>
 
<clip pos="3" side="tl" part="gen_sense_mf" link-to="3"/>
 
<clip pos="3" side="tl" part="gen_mf"/>
 
<clip pos="3" side="tl" part="nbr_sense_sp" link-to="4"/>
 
<clip pos="3" side="tl" part="nbr_sp"/>
 
<clip pos="3" side="tl" part="lemq"/>
 
</lu>
 
<b/>
 
<b pos="2"/>
 
<lu>
 
<var n="adjectiu1"/>
 
<clip pos="2" side="tl" part="lemh"/>
 
<clip pos="2" side="tl" part="a_adj"/>
 
<clip pos="2" side="tl" part="gen_sense_mf" link-to="3"/>
 
<clip pos="2" side="tl" part="gen_mf"/>
 
<clip pos="2" side="tl" part="nbr_sense_sp" link-to="4"/>
 
<clip pos="2" side="tl" part="nbr_sp" link-to="4"/>
 
<clip pos="2" side="tl" part="lemq"/>
 
</lu>
 
</chunk>
 
</out>
 
</code>
 
   
  +
The code generated by the compiler is divided in several sections. In addition, the VM reads and stores the code in its own sections.
==== Compiled Code ====
 
  +
{| class="wikitable" border="1"
  +
! Section !! Code !! VM's section !! Information
  +
|-
  +
| align=center | Header || <code>#<assembly> <br /> #<transfer default="chunk"></code> || align=center | ---- || This section establishes the type of code generated and the transfer stage.
  +
|-
  +
| align=center | Initialisation || <code>push "genere"<br /> push "<m>"<br /> storev <br /> ... <br /> jmp rules_section_start</code> || align=center | Code section || In this section we initialize the variables with their default value and execute other initialisation code. <br /> At the end we jmp to the section rules section, although rules will only execute when a pattern is matched,<br /> we need to process all the patterns which are in the rules section.
  +
|-
  +
| align=center | Macros || <code>macro_firstWord_start: <br /> ... <br /> macro_firstWord_end: <br />...</code> || align=center | Macros code section || This section contains all the macro's code delimited by labels.<br /> Each macro can be called with the 'call' instruction.
  +
|-
  +
| align=center | Patterns || <code>section_rules_start:<br /> patterns_start:<br /> push "all<predet><sp>"<br /> push "<n><pl>"<br /> push 2<br /> addtrie action_0_start<br /> ... <br />patterns_end:</code> || align=center | Preprocess code section || In this section all the patterns will be added to the system trie. <br />In this example you can see that two patterns are pushed, then the number of patterns is pushed and finally<br /> the addtrie instruction pops them and adds an entry in the trie to the rule 0.
  +
|-
  +
| align=center | Rules || <code>action_0_start:<br /> ...<br /> action_0_end:<br /> ...<br /> section_rules_end:</code> || align=center | Rules code section || Finally the rules section contains every rule delimited by its labels and all its code.
  +
|}
   
  +
*One line comments can be made by using the '#' symbol at the start of the line.
<code>
 
push "det_det_nom_adj"
 
push "<SN>"
 
pusht tipus_det ; first evaluate the variable, append/prepend '<>', then push in the stack
 
pusht gen_chunk
 
pusht nbr_chunk
 
 
push 1
 
push "^\w+" ; lem
 
cliptl
 
push 1
 
push [regex] ; a_det
 
cliptl
 
push "<3>" ; since link-to overrides everything else, we do not need any dedicated instruction
 
; for that
 
push 1
 
push [regex] ; gen_mf
 
cliptl
 
push "<4>"
 
push 1
 
push [regex] ; nbr_sp
 
cliptl
 
lu 6 ; pop 6 items, concat, create lexical unit ^...$ and push back in stack
 
 
pushbl ; push a blank
 
 
push "el"
 
push "<det><def>"
 
push "<3>"
 
push "<pl>"
 
lu 4 ; pop 4 items from the stack, create a lexical unit ^...$ and then
 
; push in the stack
 
 
pushsb 1
 
 
push 3
 
push [regex] ; lemh
 
cliptl
 
push 3
 
push [regex] ; a_nom
 
cliptl
 
push "<3>"
 
push 3
 
push [regex] ; gen_mf
 
cliptl
 
push "<4>"
 
push 3
 
push [regex] ; nbr_sp
 
cliptl
 
push 3
 
push [regex] ; lemq
 
cliptl
 
lu 7
 
 
pushbl
 
pushsb 2
 
 
pushv adjectiu1 ; its a var, so eval and push the value
 
push 3
 
push [regex] ; lemh
 
cliptl
 
push 3
 
push [regex] ; a_adj
 
cliptl
 
push "<3>"
 
push 3
 
push [regex] ; gen_mf
 
cliptl
 
push "<4>"
 
push "<4>" ; a bit confused, there are two link-to in the XML
 
push 3
 
push [regex] ; lemq
 
cliptl
 
lu 7
 
 
brace 7 ; no of blank + lexical unit = 7
 
; pop 7 items, concat, prepend and append {, } then push back
 
 
chunk 6 ; create the chunk, ^...{^...$}$, and push back in stack
 
 
out 1 ; give output (number of chunks = 1)
 
</code>
 
   
=== Example 2 ===
+
=== Code examples ===
   
==== XML t1x Code ====
+
==== Macro example ====
<code>
 
<section-def-cats>
 
<def-cat n="nom">
 
<cat-item tags="n.*"/>
 
</def-cat>
 
 
<def-cat n="det">
 
<cat-item tags="det.*"/>
 
<cat-item tags="predet.*"/>
 
</def-cat>
 
</section-def-cats>
 
 
<section-rules>
 
<rule>
 
<pattern>
 
<pattern-item n="det"/>
 
</pattern>
 
</rule>
 
<rule>
 
<pattern>
 
<pattern-item n="nom"/>
 
</pattern>
 
<action/>
 
</rule>
 
<rule>
 
<pattern>
 
<pattern-item n="det"/>
 
<pattern-item n="nom"/>
 
</pattern>
 
<action/>
 
</rule>
 
</section-rules>
 
</code>
 
   
==== Compiled Code ====
+
===== Transfer file (.t1x) =====
   
 
<code>
 
<code>
;first rule: def-cat has two equivalent cat-items
 
push "\w<det>\t" ;load pattern into stack
 
push 1
 
addtrie [address1] ;define a trie pattern with value 1 (the first rule)
 
 
push "\w<predet>\t" ;same with the second cat-item
 
push 1
 
addtrie [address1]
 
;second rule (and so on) very simple, unique cat-item
 
push "\w<n>\t"
 
push 1
 
addtrie [address2]
 
;third rule (here is the trick: multiple cat-items in one of the words)
 
push "\w<det>\t"
 
push "\w<n>\t"
 
push 2 ; we have 'det' followed by a 'nom', so addtrie has to pop two elements
 
addtrie [address3]
 
 
push "\w<predet>\t"
 
push "\w<n>\t"
 
push 2
 
addtrie [address3]
 
</code>
 
   
  +
<source lang="xml">
=== Example 3 ===
 
==== XML t1x Code ====
 
   
  +
<transfer default="chunk">
<code>
 
<def-macro n="f_coma" npar="1">
 
<choose>
 
<when>
 
<test>
 
<equal caseless="yes">
 
<clip pos="1" side="sl" part="lem"/>
 
<lit v="como"/>
 
</equal>
 
</test>
 
<let>
 
<clip pos="1" side="tl" part="lem"/>
 
<get-case-from pos="1">
 
<lit v="com a"/>
 
</get-case-from>
 
</let>
 
</when>
 
</choose>
 
</def-macro>
 
</code>
 
   
  +
<section-def-attrs>
==== Compiled code ====
 
  +
<def-attr n="nbr">
  +
<attr-item tags="sg"/>
  +
<attr-item tags="pl"/>
  +
<attr-item tags="sp"/>
  +
<attr-item tags="ND"/>
  +
</def-attr>
  +
</section-def-attrs>
   
  +
<section-def-vars>
<code>
 
  +
<def-var n="nombre" v="&amp;lt;sg&amp;gt;"/>
f_coma: push 1 ; "pos" of "clip"
 
  +
<def-var n="genere" v="&amp;lt;m&amp;gt;"/>
push "^\w+" ; "lem"
 
  +
</section-def-vars>
clipsl ; gets the value clips on the top of the stack.
 
; "sl" side is implied in the name of the instruction
 
push "como"
 
cmpi ; does the comparison and cleans the stack, it means caseless
 
jnz end ; if the comparison does not succeeds, go to end
 
; semantics: j = jump n = not z = zero flag is activated
 
; zero flag is activated when a comparison succeeds
 
; or an arithmetical operation gives 0
 
push 1 ; "pos" of "clip"
 
push "^\w+"
 
push "com a"
 
storetl ; store the value provided in the top of the stack
 
; given position 1, "tl" side and "lem"
 
 
end: ...
 
</code>
 
   
  +
<section-def-macros>
=== Example 4 ===
 
  +
<def-macro n="nombre_nom" npar="1">
==== XML t1x Code ====
 
  +
<let>
<code>
 
  +
<var n="nombre"/>
<test>
 
<or>
+
<lit v=""/>
<not>
+
</let>
<equal>
+
<choose>
<clip pos="1" side="sl" part="gen"/>
+
<when>
<clip pos="3" side="sl" part="gen"/>
+
<test>
</equal>
+
<and>
  +
<equal>
</not>
 
  +
<clip pos="1" side="sl" part="nbr"/>
<not>
 
<equal>
+
<lit-tag v="sg"/>
<clip pos="2" side="sl" part="gen"/>
+
</equal>
<clip pos="3" side="sl" part="gen"/>
+
<equal>
  +
<clip pos="1" side="tl" part="nbr"/>
</equal>
 
  +
<lit-tag v="pl"/>
</not>
 
</or>
+
</equal>
  +
</and>
</test>
 
  +
</test>
</code>
 
  +
<let>
==== Compiled code ====
 
  +
<var n="nombre"/>
 
  +
<lit-tag v="pl_slsg"/>
<code>
 
start: push 1
+
</let>
  +
</when>
push [regex] ; part="gen"
 
clipsl
 
push 3
 
push [regex] ; part="gen"
 
clipsl
 
cmp ; compare (case sensitive)
 
pushnz ; NOT zero flag and push in stack
 
 
push 2
 
push [regex] ; part="gen"
 
clipsl
 
push 3
 
push [regex] ; part="gen"
 
clipsl
 
cmp ; compare (case sensitive)
 
pushnz
 
 
or ; pop 2 items and OR, push result in stack
 
jnz end ; jump if zero flag is 0 (we did not get ZERO as the result)
 
 
... ... ...
 
(code for successful test)
 
... ... ...
 
end: ...
 
</code>
 
 
=== Example 5 ===
 
==== XML t1x Code ====
 
<def-list n="verbos_est">
 
<list-item v="actuar"/>
 
<list-item v="buscar"/>
 
<list-item v="estudiar"/>
 
<list-item v="existir"/>
 
<list-item v="ingressar"/>
 
<list-item v="introduir"/>
 
<list-item v="penetrar"/>
 
<list-item v="publicar"/>
 
<list-item v="treballar"/>
 
<list-item v="viure"/>
 
</def-list>
 
 
<rule>
 
<pattern>
 
<pattern-item n="verb"/>
 
<pattern-item n="a"/>
 
</pattern>
 
<action>
 
<choose>
 
<when>
 
<test>
 
<in caseless="yes"/>
 
<clip pos="1" side="sl" part="lem"/>
 
<list n="verbos_est"/>
 
</in>
 
</test>
 
<let>
 
<clip pos="2" side="tl" part="lem"/>
 
<lit v="en"/>
 
</let>
 
</when>
 
</choose>
 
</rule>
 
 
==== Compiled code ====
 
push "actuar"
 
push "buscar"
 
push "estudiar"
 
push "existir"
 
push "ingressar"
 
push "introduir"
 
push "penetrar"
 
push "publicar"
 
push "treballar"
 
push "viure"
 
push 10 ; number of elements in the list
 
mklist verbos_est ; make a list variable named 'verbos_est' and put the last 10 data
 
; from the stack in the list
 
 
rule1: push [regex_verb]
 
push [regex_a]
 
push 2
 
addtrie rule1_action
 
... ... ...
 
... ... ...
 
 
rule1_action: push 1
 
push "^\w+" ; lem
 
clipsl ; we have lemmma in stack now
 
incini verbox_est ; if in verbos_est (ignore case), set ZF = 1, else ZF = 0
 
jnz rule1_end
 
 
push 2
 
push "^\w+"
 
push "en"
 
storetl
 
rule1_end: ...
 
 
=== Example 6 ===
 
==== XML t1x Code ====
 
<def-macro n="firstWord" npar="1">
 
<choose>
 
<when>
 
<test>
 
<equal>
 
<clip pos="1" side="sl" part="a_np_acr"/>
 
<lit v=""/>
 
</equal>
 
</test>
 
<choose>
 
 
<when>
 
<when>
 
<test>
 
<test>
<equal>
+
<and>
<var n="EOS"/>
+
<equal>
<lit v="true"/>
+
<clip pos="1" side="sl" part="nbr"/>
</equal>
+
<lit-tag v="pl"/>
  +
</equal>
  +
<equal>
  +
<clip pos="1" side="tl" part="nbr"/>
  +
<lit-tag v="sg"/>
  +
</equal>
  +
</and>
 
</test>
 
</test>
<modify-case>
 
<clip pos="1" side="tl" part="lem"/>
 
<lit v="aa"/>
 
</modify-case>
 
 
<let>
 
<let>
<var n="caseFirstWord"/>
+
<var n="nombre"/>
<lit v="Aa"/>
+
<lit-tag v="sg_slpl"/>
 
</let>
 
</let>
 
</when>
 
</when>
<otherwise>
+
<otherwise>
 
<let>
 
<let>
<var n="caseFirstWord"/>
+
<var n="nombre"/>
<lit v="aa"/>
+
<clip pos="1" side="tl" part="nbr"/>
 
</let>
 
</let>
 
</otherwise>
 
</otherwise>
</choose>
+
</choose>
</when>
+
</def-macro>
  +
</section-def-macros>
<otherwise>
 
  +
<let>
 
  +
</transfer>
<var n="caseFirstWord"/>
 
  +
<lit v="aa"/>
 
  +
</source>
</let>
 
  +
</otherwise>
 
  +
</code>
</choose>
 
  +
<let>
 
  +
===== Code generated =====
<var n="EOS"/>
 
  +
<lit v="false"/>
 
  +
<source lang="bash">
</let>
 
</def-macro>
+
#<assembly>
  +
#<transfer default="chunk">
 
  +
#<def-var v="&lt;sg&gt;" n="nombre">
 
  +
push "nombre"
<rule comment="REGLA: DET DET ADJ NOM (your many beautiful cats)">
 
  +
push "<sg>"
... ...
 
  +
storev
<action>
 
  +
#<def-var v="&lt;m&gt;" n="genere">
<call-macro n="firstWord">
 
  +
push "genere"
<with-param pos="1"/>
 
  +
push "<m>"
</call-macro>
 
  +
storev
<call-macro n="f_concord4">
 
  +
jmp section_rules_start
<with-param pos="4"/>
 
  +
#<def-macro npar="1" n="nombre_nom">
<with-param pos="3"/>
 
  +
macro_nombre_nom_start:
<with-param pos="2"/>
 
  +
#<var n="nombre">
<with-param pos="1"/>
 
  +
push "nombre"
</call-macro>
 
  +
#<lit v="">
...
 
  +
push ""
<out>
 
  +
storev
<chunk name="det_det_nom_adj" case="caseFirstWord">
 
  +
#<clip part="nbr" pos="1" side="sl">
... ...
 
  +
push 1
</chunk>
 
  +
push "<sg>|<pl>|<sp>|<ND>"
</out>
 
  +
clipsl
</action>
 
  +
#<lit-tag v="sg">
</rule>
 
  +
push "<sg>"
  +
cmp
  +
#<clip part="nbr" pos="1" side="tl">
  +
push 1
  +
push "<sg>|<pl>|<sp>|<ND>"
  +
cliptl
  +
#<lit-tag v="pl">
  +
push "<pl>"
  +
cmp
  +
and 2
  +
jz when_0_end
  +
#<var n="nombre">
  +
push "nombre"
  +
#<lit-tag v="pl_slsg">
  +
push "<pl_slsg>"
  +
storev
  +
jmp choose_0_end
  +
when_0_end:
  +
#<clip part="nbr" pos="1" side="sl">
  +
push 1
  +
push "<sg>|<pl>|<sp>|<ND>"
  +
clipsl
  +
#<lit-tag v="pl">
  +
push "<pl>"
  +
cmp
  +
#<clip part="nbr" pos="1" side="tl">
  +
push 1
  +
push "<sg>|<pl>|<sp>|<ND>"
  +
cliptl
  +
#<lit-tag v="sg">
  +
push "<sg>"
  +
cmp
  +
and 2
  +
jz when_1_end
  +
#<var n="nombre">
  +
push "nombre"
  +
#<lit-tag v="sg_slpl">
  +
push "<sg_slpl>"
  +
storev
  +
jmp choose_0_end
  +
when_1_end:
  +
#<otherwise>
  +
#<var n="nombre">
  +
push "nombre"
  +
#<clip part="nbr" pos="1" side="tl">
  +
push 1
  +
push "<sg>|<pl>|<sp>|<ND>"
  +
cliptl
  +
storev
  +
choose_0_end:
  +
macro_nombre_nom_end: ret
  +
#<section-rules>
  +
section_rules_start:
  +
section_rules_end:
  +
</source>
  +
  +
----
  +
  +
==== Rule's patterns example ====
  +
  +
===== Transfer file (.t1x) =====
  +
<source lang="xml">
  +
<transfer default="chunk">
  +
  +
<section-def-cats>
  +
<def-cat n="all">
  +
<cat-item lemma="all" tags="predet.sp"/>
  +
</def-cat>
  +
  +
<def-cat n="adj2">
  +
<cat-item tags="adj"/>
  +
<cat-item tags="adj.*"/>
  +
<cat-item tags="adj.sint"/>
  +
<cat-item tags="adj.sint.*"/>
  +
<cat-item tags="adj.comp"/>
  +
<cat-item tags="adj.sup"/>
  +
</def-cat>
  +
  +
<def-cat n="nomcomu">
  +
<cat-item tags="n.*"/>
  +
</def-cat>
  +
  +
<def-cat n="nompropi">
  +
<cat-item tags="np.*"/>
  +
</def-cat>
  +
  +
<def-cat n="nploc">
  +
<cat-item tags="np.loc.*"/>
  +
</def-cat>
  +
</section-def-cats>
  +
  +
<rule>
  +
<pattern>
  +
<pattern-item n="all"/>
  +
<pattern-item n="adj2"/>
  +
</pattern>
  +
<action>
  +
...
  +
</action>
  +
</rule>
  +
  +
<rule>
  +
<pattern>
  +
<pattern-item n="nomcomu"/>
  +
<pattern-item n="nompropi"/>
  +
<pattern-item n="nploc"/>
  +
</pattern>
  +
<action>
  +
...
  +
</action>
  +
</rule>
  +
</section-rules>
  +
  +
</transfer>
  +
  +
</source>
  +
  +
===== Code generated =====
  +
  +
<source lang="bash">
  +
#<assembly>
  +
#<transfer default="chunk">
  +
jmp section_rules_start
  +
#<section-rules>
  +
section_rules_start:
  +
patterns_start:
  +
push "all<predet><sp>"
  +
push "<adj>|<adj><*>|<adj><sint>|<adj><sint><*>|<adj><comp>|<adj><sup>"
  +
push 2
  +
addtrie action_0_start
  +
push "<n><*>"
  +
push "<np><*>"
  +
push "<np><loc><*>"
  +
push 3
  +
addtrie action_1_start
  +
patterns_end:
  +
action_0_start:
  +
...
  +
action_0_end:
  +
action_1_start:
  +
...
  +
action_1_end:
  +
section_rules_end:
  +
</source>
  +
  +
----
  +
   
  +
==Wishlist==
==== Compiled code ====
 
   
  +
* Make the compile options like with lt-comp/cg-comp (e.g. you don't need -i/-o to say what the input/output files are), just <code>apertium-compiler [options] <infile> <outfile></code>
firstWord:
 
... ... ; normal translation of instructions, all the variables are assumed global
 
... ...
 
ret ; ret instruction does a number of things
 
; pops 'frame stack', current 'local variable frame' is reset with popped
 
; values (actually its more pointer assignment), C++ version will also
 
; do the necessary deallocations
 
; pops global stack, update PC with the popped value
 
... ...
 
... ...
 
rule_ddan_action: push 1 ; pos = 1
 
push 1 ; number of parameters 1
 
call firstWord ; macro label
 
; call statement does a number of things
 
; 1. temppc = PC + 1, set PC = firstWord
 
; 2. pushes the current 'local variable frame' into 'frame stack'
 
; 3. create a new 'local variable frame'
 
; 4. pops the arguments from the stack and places then in the 'local
 
; variable frame'
 
; 5. pushes temppc in global stack (it will be used by the return
 
; statement)
 
 
; 6. continue (instruction at firstWord will be evaluated next)
 
push 1 ; notice that the arguments are pushed in reverse order
 
; when popped, they will be in the right order
 
push 2
 
push 3
 
push 4
 
push 4
 
call f_concord4
 
... ...
 
   
  +
==External links==
== Development Notes ==
 
   
  +
* [https://github.com/ggm/vm-for-transfer-cpp Github: vm-for-transfer-cpp]
* None of the macro and actions need to return anything (unlike conventional functions), so provision for returning a value (using stack) is unnecessary
 
   
  +
[[Category:Development]]
* The local variable frame is actually a queue with a maximum length equal to the maximum pattern length in the trie.
 
  +
[[Category:Documentation in English]]
  +
[[Category:Transfer]]

Latest revision as of 13:58, 7 October 2014

En français

Instruction Set[edit]

Mnemonic Opcode
(in hex)
Other operands Stack
[before]→[after] (top, top-1, ...)
Description
push - value [empty] → value Pushes a string or a variable value onto the stack. Strings go between quotes ("string") but variable's names not
pushbl - N/A [empty] → blank Pushes a blank onto the stack
pushsb - pos [empty] → superblank Pushes the superblank at 'pos' onto the stack
append - N valueN, ..., value1, varName → [empty] Pops 'N' elements and appends them to a variable or clip
concat - N valueN, ..., value1 → value1...valueN Pops 'N' elements and pushes them back concatenated
clip - N/A part → value Obtains the part in the only language there is (inter/post-chunk) and pushes the value onto the stack
clipsl - link-to part, pos → value Obtains the 'part' in source language in position 'pos' and pushes the 'value' onto the stack. An optional operand is used for clips with link-to tags, e.g. "clipsl <3>".
cliptl - link-to part, pos → value Obtains the 'part' in target language in position 'pos' and pushes the 'value' onto the stack. An optional operand is used for clips with link-to tags, e.g. "cliptl <3>".
storecl - N/A value, part → [empty] Stores 'value' in the only language there is (inter/post-chunk)
storesl - N/A value, part, pos → [empty] Stores 'value' as the 'part' of the source language in position 'pos'
storetl - N/A value, part, pos → [empty] Stores 'value' as the 'part' of the target language in position 'pos'
storev - N/A value, varName → [empty] Stores 'value' in the variable with name 'varName'
addtrie - address N, patternN, ..., pattern1 → [empty] Pops 'N' patterns and creates a trie entry pointing to 'address'
lu - N valueN, ..., value1 → ^(lexical_unit)$ Pops 'N' values from the stack, creates a lexical unit ^...$ with them and pushes the lu back onto the stack
mlu - N luN, ..., lu1 → multiword Pops 'N' lu from the stack, creates a multiword with them and pushes the multiword back onto the stack
lu-count - N/A [empty] → number Pushes the number of lexical units (words inside the chunk) in the rule onto the stack
chunk - N elemN-2, ... , elem1, <tags>, name → ^name<tags>{elem1...elemN-2}$ Pops 'N' amount of data from the stack, creates the chunk and pushes it back onto the stack
out - N valueN, ..., value1 → [empty] Pops 'N' values from the stack and outputs them
cmp - N/A value2, value1 → result Pops 'value1' and 'value2', compares them, if they are equal pushes a 1 (true), if they aren't pushes a 0 (false)
cmpi - N/A value2, value1 → result Pops 'value1' and 'value2', compares them (ignoring case for each string), if they are equal pushes a 1 (true), if they aren't pushes a 0 (false)
cmp-substr - N/A value2, value1 → result Tests if 'value1' contains the substring 'value2', result can be 1 (true) or 0 (false).
cmpi-substr - N/A value2, value1 → result Tests if 'value1' contains the substring 'value2' (ignoring case for each string), result can be 1 (true) or 0 (false).
not - N value → result Negates the value on top of the stack, 0 -> 1 or 1 -> 0
and - N valueN, ..., value1 → result And operation of 'N' values, result can be 1 (true) or 0 (false)
or - N valueN, ..., value1 → result Or operation of 'N' values, result can be 1 (true) or 0 (false)
in - N/A list, value → result Performs a search of a 'value' in a 'list'
inig - N/A list, value → result Performs a search (ignoring case) of a 'value' in a 'list'
jmp - label [empty] → [empty] Jumps to the label, unconditionally
jz - label top → [empty] Jumps to the label if stack.top == 0
jnz - label top → [empty] Jumps to the label if stack.top == 1
call - label N, argN, ..., arg1 → [empty] Calls a macro with the arguments on the stack
ret - N/A [empty] → [empty] Returns from a macro, PC will be handled automatically by the VM.
nop - N/A [empty] → [empty] No operation
case-of - N/A container → case Gets the case from the container in the stack. The container would usually be the result of a clip instruction but can be any string.
get-case-from - N/A pos → case Gets the case from the lexical unit in position 'pos'
modify-case - N/A case, container → modifiedContainer Modifies the case of the 'container' to 'case' and leaves the modified container on the stack
begins-with - N/A value2, value1 → result Checks if 'value1' begins with 'value2' and pushes 1 (true) or 0 (false), 'value2' can be a list
begins-with-ig - N/A value2, value1 → result Checks if 'value1' begins with 'value2' (ignoring the case) and pushes 1 (true) or 0 (false), 'value2' can be a list
ends-with - N/A value2, value1 → result Checks if 'value1' ends with 'value2' and pushes 1 (true) or 0 (false), 'value2' can be a list
ends-with-ig - N/A value2, value1 → result Checks if 'value1' ends with 'value2' (ignoring the case) and pushes 1 (true) or 0 (false), 'value2' can be a list
  • Lists are represented as a concatenation of items separated by '|', e.g. uno|otro|poco|cuánto|menos|mucho|tanto|demasiado
  • The case is represented as "aa" (all lowercase), "Aa" (first uppercase) and "AA", (all uppercase).

Code generation[edit]

Code sections[edit]

The code generated by the compiler is divided in several sections. In addition, the VM reads and stores the code in its own sections.

Section Code VM's section Information
Header #<assembly>
#<transfer default="chunk">
---- This section establishes the type of code generated and the transfer stage.
Initialisation push "genere"
push "<m>"
storev
...
jmp rules_section_start
Code section In this section we initialize the variables with their default value and execute other initialisation code.
At the end we jmp to the section rules section, although rules will only execute when a pattern is matched,
we need to process all the patterns which are in the rules section.
Macros macro_firstWord_start:
...
macro_firstWord_end:
...
Macros code section This section contains all the macro's code delimited by labels.
Each macro can be called with the 'call' instruction.
Patterns section_rules_start:
patterns_start:
push "all<predet><sp>"
push "<n><pl>"
push 2
addtrie action_0_start
...
patterns_end:
Preprocess code section In this section all the patterns will be added to the system trie.
In this example you can see that two patterns are pushed, then the number of patterns is pushed and finally
the addtrie instruction pops them and adds an entry in the trie to the rule 0.
Rules action_0_start:
...
action_0_end:
...
section_rules_end:
Rules code section Finally the rules section contains every rule delimited by its labels and all its code.
  • One line comments can be made by using the '#' symbol at the start of the line.

Code examples[edit]

Macro example[edit]

Transfer file (.t1x)[edit]

 <transfer default="chunk">

  <section-def-attrs>
    <def-attr n="nbr">
      <attr-item tags="sg"/>
      <attr-item tags="pl"/>
      <attr-item tags="sp"/>
      <attr-item tags="ND"/>
    </def-attr>
  </section-def-attrs>

  <section-def-vars>
    <def-var n="nombre" v="&amp;lt;sg&amp;gt;"/>
    <def-var n="genere" v="&amp;lt;m&amp;gt;"/>
  </section-def-vars>

  <section-def-macros>
    <def-macro n="nombre_nom" npar="1">
      <let>
        <var n="nombre"/>
        <lit v=""/>
      </let>
      <choose>
        <when>
          <test>
            <and>
              <equal>         
                <clip pos="1" side="sl" part="nbr"/>
                <lit-tag v="sg"/>
              </equal>
              <equal>
                <clip pos="1" side="tl" part="nbr"/>
                <lit-tag v="pl"/>
              </equal>
            </and>
          </test>
          <let>
            <var n="nombre"/>
            <lit-tag v="pl_slsg"/>
          </let>
         </when>
         <when>
           <test>
             <and>
               <equal>
                 <clip pos="1" side="sl" part="nbr"/>
                 <lit-tag v="pl"/>
               </equal>
               <equal>
                 <clip pos="1" side="tl" part="nbr"/>
                 <lit-tag v="sg"/>
               </equal>
             </and>
           </test>
           <let>
             <var n="nombre"/>
             <lit-tag v="sg_slpl"/>
           </let>
         </when>
         <otherwise>
           <let>
             <var n="nombre"/>
             <clip pos="1" side="tl" part="nbr"/>
           </let>
         </otherwise>
       </choose> 
     </def-macro>
   </section-def-macros>

 </transfer>

Code generated[edit]
 #<assembly>
 #<transfer default="chunk">
 #<def-var v="&lt;sg&gt;" n="nombre">
 push "nombre"
 push "<sg>"
 storev
 #<def-var v="&lt;m&gt;" n="genere">
 push "genere"
 push "<m>"
 storev
 jmp section_rules_start
 #<def-macro npar="1" n="nombre_nom">
 macro_nombre_nom_start:
 #<var n="nombre">
 push "nombre"
 #<lit v="">
 push ""
 storev
 #<clip part="nbr" pos="1" side="sl">
 push 1
 push "<sg>|<pl>|<sp>|<ND>"
 clipsl
 #<lit-tag v="sg">
 push "<sg>"
 cmp
 #<clip part="nbr" pos="1" side="tl">
 push 1
 push "<sg>|<pl>|<sp>|<ND>"
 cliptl
 #<lit-tag v="pl">
 push "<pl>"
 cmp
 and 2
 jz when_0_end
 #<var n="nombre">
 push "nombre"
 #<lit-tag v="pl_slsg">
 push "<pl_slsg>"
 storev
 jmp choose_0_end
 when_0_end:
 #<clip part="nbr" pos="1" side="sl">
 push 1
 push "<sg>|<pl>|<sp>|<ND>"
 clipsl
 #<lit-tag v="pl">
 push "<pl>"
 cmp
 #<clip part="nbr" pos="1" side="tl">
 push 1
 push "<sg>|<pl>|<sp>|<ND>"
 cliptl
 #<lit-tag v="sg">
 push "<sg>"
 cmp
 and 2
 jz when_1_end
 #<var n="nombre">
 push "nombre"
 #<lit-tag v="sg_slpl">
 push "<sg_slpl>"
 storev
 jmp choose_0_end
 when_1_end:
 #<otherwise>
 #<var n="nombre">
 push "nombre"
 #<clip part="nbr" pos="1" side="tl">
 push 1
 push "<sg>|<pl>|<sp>|<ND>"
 cliptl
 storev
 choose_0_end:
 macro_nombre_nom_end: ret
 #<section-rules>
 section_rules_start:
 section_rules_end:

Rule's patterns example[edit]

Transfer file (.t1x)[edit]
<transfer default="chunk">

  <section-def-cats>
    <def-cat n="all">
      <cat-item lemma="all" tags="predet.sp"/>
    </def-cat>

    <def-cat n="adj2">
      <cat-item tags="adj"/>
      <cat-item tags="adj.*"/>
      <cat-item tags="adj.sint"/>
      <cat-item tags="adj.sint.*"/>
      <cat-item tags="adj.comp"/>
      <cat-item tags="adj.sup"/>
    </def-cat>

    <def-cat n="nomcomu">
      <cat-item tags="n.*"/>
    </def-cat>
    
    <def-cat n="nompropi">
      <cat-item tags="np.*"/>
    </def-cat>

    <def-cat n="nploc">
      <cat-item tags="np.loc.*"/>
    </def-cat>
  </section-def-cats>

    <rule> 
      <pattern>
	<pattern-item n="all"/>
	<pattern-item n="adj2"/>
      </pattern>
      <action>
	...
      </action>
    </rule>

    <rule> 
      <pattern>
	<pattern-item n="nomcomu"/>
	<pattern-item n="nompropi"/>
	<pattern-item n="nploc"/>
      </pattern>
      <action>
	...
      </action>
    </rule>
  </section-rules>

</transfer>
Code generated[edit]
#<assembly>
#<transfer default="chunk">
jmp section_rules_start
#<section-rules>
section_rules_start:
patterns_start:
push "all<predet><sp>"
push "<adj>|<adj><*>|<adj><sint>|<adj><sint><*>|<adj><comp>|<adj><sup>"
push 2
addtrie action_0_start
push "<n><*>"
push "<np><*>"
push "<np><loc><*>"
push 3
addtrie action_1_start
patterns_end:
action_0_start:
...
action_0_end:
action_1_start:
...
action_1_end:
section_rules_end:


Wishlist[edit]

  • Make the compile options like with lt-comp/cg-comp (e.g. you don't need -i/-o to say what the input/output files are), just apertium-compiler [options] <infile> <outfile>

External links[edit]