Skip to content

Latest commit

 

History

History
838 lines (749 loc) · 25.4 KB

xml-protocol.md

File metadata and controls

838 lines (749 loc) · 25.4 KB
Raw

Coq XML Protocol

This document is based on documentation originally written by CJ Bell for his vscoq project.

Here, the aim is to provide a "hands on" description of the XML protocol that coqtop and IDEs use to communicate. The protocol first appeared with Coq 8.5, and is used by CoqIDE, vscoq, and other user interfaces.

A somewhat out-of-date description of the async state machine is documented here. OCaml types for the protocol can be found in the ide/protocol/interface.ml file.

Changes to the XML protocol are documented as part of dev/doc/changes.md.

Sentences: each command sent to Coqtop is a "sentence"; they are typically terminated by ".\s" (followed by whitespace or EOF). Examples: "Lemma a: True.", "(* asdf *) Qed.", "auto; reflexivity." In practice, the command sentences sent to Coqtop are terminated at the "." and start with any previous whitespace. Each sentence is assigned a unique stateId after being sent to Coq (via Add). States:

  • Processing: has been received by Coq and has no obvious syntax error (that would prevent future parsing)
  • Processed:
  • InProgress:
  • Incomplete: the validity of the sentence cannot be checked due to a prior error
  • Complete:
  • Error: the sentence has an error

State ID 0 is reserved as a 'dummy' state.

Commands

About(unit)

Returns information about the protocol and build dates for Coqtop.

<call val="About">
  <unit/>
</call>

Returns

 <value val="good">
   <coq_info><string>8.6</string>
     <string>20150913</string>
     <string>December 2016</string>
     <string>Dec 23 2016 16:16:30</string>
   </coq_info>
</value>

The string fields are the Coq version, the protocol version, the release date, and the compile time of Coqtop. The protocol version is a date in YYYYMMDD format, where "20150913" corresponds to Coq 8.6. An IDE that wishes to support multiple Coq versions can use the protocol version information to know how to handle output from Coqtop.

Add(stateId: integer, command: string, verbose: boolean)

Adds a toplevel command (e.g. vernacular, definition, tactic) to the given state. verbose controls whether out-of-band messages will be generated for the added command (e.g. "foo is assumed" in response to adding "Axiom foo: nat.").

<call val="Add">
  <pair>
    <pair>
      <string>${command}</string>
      <int>${editId}</int>
    </pair>
    <pair>
      <state_id val="${stateId}"/>
      <bool val="${verbose}"/>
    </pair>
  </pair>
</call>

Returns

  • The added command is given a fresh stateId and becomes the next "tip".
<value val="good">
  <pair>
    <state_id val="${newStateId}"/>
    <pair>
      <union val="in_l"><unit/></union>
      <string>${message}</string>
    </pair>
  </pair>
</value>
  • When closing a focused proof (in the middle of a bunch of interpreted commands), the Qed will be assigned a prior stateId and nextStateId will be the id of an already-interpreted state that should become the next tip.
<value val="good">
  <pair>
    <state_id val="${stateId}"/>
    <pair>
      <union val="in_r"><state_id val="${nextStateId}"/></union>
      <string>${message}</string>
    </pair>
  </pair>
</value>
  • Failure:
    • Syntax error. Error offsets are byte offsets (not character offsets) with respect to the start of the sentence, starting at 0.
    <value val="fail"
    loc_s="${startOffsetOfError}"
    loc_e="${endOffsetOfError}">
  <state_id val="${stateId}"/>
  <richpp>${errorMessage}</richpp>
</value>
    • Another kind of error, for example, Qed with a pending goal.
    <value val="fail"><state_id val="${stateId}"/><richpp>${errorMessage}</richpp></value>

EditAt(stateId: integer)

Moves current tip to ${stateId}, such that commands may be added to the new state ID.

<call val="Edit_at"><state_id val="${stateId}"/></call>

Returns

  • Simple backtrack; focused stateId becomes the parent state
<value val="good">
  <union val="in_l"><unit/></union>
</value>
  • New focus; focusedQedStateId is the closing Qed of the new focus; senteneces between the two should be cleared
<value val="good">
  <union val="in_r">
    <pair>
      <state_id val="${focusedStateId}"/>
      <pair>
        <state_id val="${focusedQedStateId}"/>
        <state_id val="${oldFocusedStateId}"/>
      </pair>
    </pair>
  </union>
</value>
  • Failure: If stateId is in an error-state and cannot be jumped to, errorFreeStateId is the parent state of stateId that should be edited instead.
<value val="fail" loc_s="${startOffsetOfError}" loc_e="${endOffsetOfError}">
  <state_id val="${errorFreeStateId}"/>
  ${errorMessage}
</value>

Init()

  • No options.
<call val="Init"><option val="none"/></call>
  • With options. Looking at ide_slave.ml, it seems that options is just the name of a script file, whose path is added via Add LoadPath to the initial state.
<call val="Init">
  <option val="some">
    <string>${options}</string>
  </option>
</call>

Providing the script file enables Coq to use .aux files created during compilation. Those file contain timing information that allow Coq to choose smartly between asynchronous and synchronous processing of proofs.

Returns

  • The initial stateId (not associated with a sentence)
<value val="good">
  <state_id val="${initialStateId}"/>
</value>

Goal()

<call val="Goal"><unit/></call>

Returns

  • If there is a goal. shelvedGoals and abandonedGoals have the same structure as the first set of (current/foreground) goals. backgroundGoals contains a list of pairs of lists of goals (list ((list Goal)*(list Goal))); it represents a "focus stack" (see code for reference). Each time a proof is focused, it will add a new pair of lists-of-goals. The first pair is the most nested set of background goals, the last pair is the top level set of background goals. The first list in the pair is in reverse order. Each time you focus the goal (e.g. using Focus or a bullet), a new pair will be prefixed to the list.
<value val="good">
  <option val="some">
  <goals>
    <!-- current goals -->
    <list>
      <goal>
        <string>3</string>
        <list>
          <richpp>${hyp1}</richpp>
          ...
          <richpp>${hypN}</richpp>
        </list>
        <richpp>${goal}</richpp>
      </goal>
      ...
      ${goalN}
    </list>
    <!-- `backgroundGoals` -->
    <list>
      <pair>
        <list><goal />...</list>
        <list><goal />...</list>
      </pair>
      ...
    </list>
    ${shelvedGoals}
    ${abandonedGoals}
  </goals>
  </option>
</value>

For example, this script:

Goal P -> (1=1/\2=2) /\ (3=3 /\ (4=4 /\ 5=5) /\ 6=6) /\ 7=7.
intros.
split; split. (* current visible goals are [1=1, 2=2, 3=3/\(4=4/\5=5)/\6=6, 7=7] *)
Focus 3. (* focus on 3=3/\(4=4/\5=5)/\6=6; bg-before: [1=1, 2=2], bg-after: [7=7] *)
split; [ | split ]. (* current visible goals are [3=3, 4=4/\5=5, 6=6] *)
Focus 2. (* focus on 4=4/\5=5; bg-before: [3=3], bg-after: [6=6] *)
* (* focus again on 4=4/\5=5; bg-before: [], bg-after: [] *)
split. (* current visible goals are [4=4,5=5] *)

should generate the following goals structure:

goals: [ P|-4=4, P|-5=5 ]
background:
[
  ( [], [] ), (* bullet with one goal has no before or after background goals *)
  ( [ P|-3=3 ], [ P|-6=6 ] ), (* Focus 2 *)
  ( [ P|-2=2, P|-1=1 ], [ P|-7=7 ] ) (* Focus 3; notice that 1=1 and 2=2 are reversed *)
]

Pseudocode for listing all of the goals in order: rev (flat_map fst background) ++ goals ++ flat_map snd background.

  • No goal:
<value val="good"><option val="none"/></value>

Status(force: bool)

Returns information about the current proofs. CoqIDE typically sends this message with force = false after each sentence, and with force = true if the user wants to force the checking of all proofs (wheels button). In terms of the STM API, force triggers a Join.

<call val="Status"><bool val="${force}"/></call>

Returns

<status>
  <string>${path}</string>
  <string>${proofName}</string>
  <string>${allProofs}</string>
  <string>${proofNumber}</string>
</status>

Query(route_id: integer, query: string, stateId: integer)

routeId can be used to distinguish the result of a particular query, stateId should be set to the state the query should be run.

<call val="Query">
  <pair>
    <route_id val="${routeId}"/>
  <pair>
    <string>${query}</string>
    <state_id val="${stateId}"/>
  </pair>
  </pair>
</call>

Returns

<value val="good">
  <string>${message}</string>
</value>

Before 8.8, Query only executed the first command present in the query string; starting with 8.8, the caller may include several statements. This is useful for instance for temporarily setting an option and then executing a command.

Evars()

<call val="Evars"><unit/></call>

Returns

<value val="good">
  <option val="some">
    <list>
      <evar>${evar1}</evar>
      ...
      <evar>${evarN}</evar>
    </list>
  </option>
</value>

Hints()

<call val="Hints"><unit/></call>

Returns

<value val="good">
  <option val="some">
    <pair>
      <list/>
      <list>
        <pair>
          <string>${hint1}</string>
          <string>${hint2}</string>
        </pair>
        ...
        <pair>
          <string>${hintN-1}</string>
          <string>${hintN}</string>
        </pair>
      </list>
    </pair>
  </option>
</value>

Search([(constraintTypeN: string, constraintValueN: string, positiveConstraintN: boolean)])

Searches for objects that satisfy a list of constraints. If ${positiveConstraint} is false, then the constraint is inverted.

<call val="Search">
  <list>
    <pair>
      <search_cst val="${constraintType1}">
        ${constraintValue1}
      </search_cst>
      <bool val="${positiveConstraint1}"/>
    </pair>
    ...
    <!-- Example: -->
    <pair>
      <search_cst val="name_pattern">
        <string>bool_rect</string>
      </search_cst>
      <bool val="true"/>
    </pair>
  </list>
</call>

Returns

<value val="good">
  <list>
      <coq_object>
          <list>
              <string>${metaInfo}</string>
              ...
          </list>
          <list>
              <string>${name}</string>
          </list>
          <string>${definition}</string>
      </coq_object>
      ...
  </list>
</value>
Types of constraints:
  • Name pattern: ${constraintType} = "name_pattern"; ${constraintValue} is a regular expression string.
  • Type pattern: ${constraintType} = "type_pattern"; ${constraintValue} is a pattern (???: an open gallina term) string.
  • SubType pattern: ${constraintType} = "subtype_pattern"; ${constraintValue} is a pattern (???: an open gallina term) string.
  • In module: ${constraintType} = "in_module"; ${constraintValue} is a list of strings specifying the module/directory structure.
  • Include blacklist: ${constraintType} = "include_blacklist"; ${constraintValue} is omitted.

GetOptions()

<call val="GetOptions"><unit/></call>

Returns

<value val="good">
  <list>
    <pair>
      <list><string>${string1}</string>...</list>
      <option_state>
        <bool>${sync}</bool>
        <bool>${deprecated}</bool>
        <string>${name}</string>
        ${option_value}
      </option_state>
    </pair>
    ...
  </list>
</value>

SetOptions(options)

Sends a list of option settings, where each setting roughly looks like: ([optionNamePart1, ..., optionNamePartN], value).

<call val="SetOptions">
  <list>
    <pair>
      <list>
        <string>optionNamePart1</string>
        ...
        <string>optionNamePartN</string>
      </list>
      <option_value val="${typeOfOption}">
        <option val="some">
          ${value}
        </option>
      </option_value>
    </pair>
    ...
    <!-- Example: -->
    <pair>
      <list>
        <string>Printing</string>
        <string>Width</string>
      </list>
      <option_value val="intvalue">
        <option val="some"><int>60</int></option>
      </option_value>
    </pair>
  </list>
</call>

CoqIDE sends the following settings (defaults in parentheses):

Printing Width : (<option_value val="intvalue"><int>60</int></option_value>),
Printing Coercions : (<option_value val="boolvalue"><bool val="false"/></option_value>),
Printing Matching : (...true...)
Printing Notations : (...true...)
Printing Existential Instances : (...false...)
Printing Implicit : (...false...)
Printing All : (...false...)
Printing Universes : (...false...)

Returns

<value val="good"><unit/></value>

MkCases(...)

<call val="MkCases"><string>...</string></call>

Returns

<value val="good">
  <list>
    <list><string>${string1}</string>...</list>
    ...
  </list>
</value>

StopWorker(worker: string)

<call val="StopWorker"><string>${worker}</string></call>

Returns

<value val="good"><unit/></value>

PrintAst(stateId: integer)

<call val="PrintAst"><state_id val="${stateId}"/></call>

Returns

<value val="good">
  <gallina begin="${gallina_begin}" end="${gallina_end}">
    <theorem begin="${theorem_begin}" end="${theorem_end}" type="Theorem" name="${theorem_name}">
      <apply begin="${apply_begin}" end="${apply_end}">
        <operator begin="${operator_begin}" end="${operator_end}" name="${operator_name}"/>
        <typed begin="${typed_begin}" end="${typed_end}">
          <constant begin="${constant_begin}" end="${constant_end}" name="${constant_name}"/>
          ...
          <token begin="${token_begin}" end="token_end">${token}</token>
          ...
        </typed>
        ...
      </apply>
    </theorem>
    ...
  </gallina>
</value>

Annotate(annotation: string)

<call val="Annotate"><string>${annotation}</string></call>

Returns

take <call val="Annotate"><string>Theorem plus_0_r : forall n : nat, n + 0 = n.</string></call> as an example.

<value val="good">
  <pp startpos="0" endpos="45">
    <vernac_expr startpos="0" endpos="44">
      <keyword startpos="0" endpos="7">Theorem</keyword>
      &nbsp;plus_0_r&nbsp;:&nbsp;
      <constr_expr startpos="19" endpos="44">
        <keyword startpos="19" endpos="25">forall</keyword>
        &nbsp;n&nbsp;:&nbsp;
        <constr_expr startpos="30" endpos="33">nat</constr_expr>
        ,&nbsp;
        <unparsing startpos="35" endpos="44">
          <unparsing startpos="35" endpos="40">
            <unparsing startpos="35" endpos="40">
              <unparsing startpos="35" endpos="36">
                <constr_expr startpos="35" endpos="36">n</constr_expr>
              </unparsing>
              <unparsing startpos="36" endpos="38">&nbsp;+</unparsing>
              <unparsing startpos="38" endpos="39">&nbsp;</unparsing>
              <unparsing startpos="39" endpos="40">
                <constr_expr startpos="39" endpos="40">0</constr_expr>
              </unparsing>
            </unparsing>
          </unparsing>
          <unparsing startpos="40" endpos="42">&nbsp;=</unparsing>
          <unparsing startpos="42" endpos="43">&nbsp;</unparsing>
          <unparsing startpos="43" endpos="44">
            <constr_expr startpos="43" endpos="44">n</constr_expr>
          </unparsing>
        </unparsing>
      </constr_expr>
    </vernac_expr>
    .
  </pp>
</value>

Db_cmd(user_input: string)

<call val="Db_cmd"><string>${user_input}</string></call>

Returns

<call val="Db_cmd"><string>h</string></call> passes the command "h" to the debugger. It returns unit.

Feedback messages

Feedback messages are issued out-of-band, giving updates on the current state of sentences/stateIds, worker-thread status, etc.

In the descriptions of feedback syntax below, wherever a state_id tag may occur, there may instead be an edit_id tag.

  • Added Axiom: in response to Axiom, admit, Admitted, etc.
<feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="addedaxiom" />
</feedback>
<feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="processingin">
    <string>${workerName}</string>
  </feedback_content>
</feedback>
<feedback object="state" route="0">
  <feedback object="state" route="0">
    <state_id val="${stateId}"/>
  <feedback_content val="processed"/>
</feedback>
<feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="incomplete" />
</feedback>
  • Complete
  • GlobRef
  • Error. Issued, for example, when a processed tactic has failed or is unknown. The error offsets may both be 0 if there is no particular syntax involved.
  • InProgress
<feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="inprogress">
    <int>1</int>
  </feedback_content>
</feedback>
  • WorkerStatus Ex: workername = "proofworker:0" Ex: status = "Idle" or status = "proof: myLemmaName" or status = "Dead"
<feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="workerstatus">
    <pair>
      <string>${workerName}</string>
      <string>${status}</string>
    </pair>
  </feedback_content>
</feedback>
  • File Dependencies. Typically in response to a Require. Dependencies are *.vo files.
    • State stateId directly depends on dependency:
    <feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="filedependency">
    <option val="none"/>
    <string>${dependency}</string>
  </feedback_content>
</feedback>
    • State stateId depends on dependency via dependency sourceDependency
    <feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="filedependency">
    <option val="some"><string>${sourceDependency}</string></option>
    <string>${dependency}</string>
  </feedback_content>
</feedback>
  • File Loaded. For state stateId, module module is being loaded from voFileName
<feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="fileloaded">
    <string>${module}</string>
    <string>${voFileName`}</string>
  </feedback_content>
</feedback>
  • Message. level is one of {info,warning,notice,error,debug}. For example, in response to an add "Axiom foo: nat." with verbose=true, message foo is assumed will be emitted in response.
<feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="message">
    <message>
      <message_level val="${level}"/>
      <string>${message}</string>
    </message>
  </feedback_content>
</feedback>
  • Custom. A feedback message that Coq plugins can use to return structured results, including results from Ltac profiling. Optionally, startPos and stopPos define a range of offsets in the document that the message refers to; otherwise, they will be 0. customTag is intended as a unique string that identifies what kind of payload is contained in customXML.
<feedback object="state" route="0">
  <state_id val="${stateId}"/>
  <feedback_content val="custom">
    <loc start="${startPos}" stop="${stopPos}"/>
    <string>${customTag}</string>
    ${customXML}
  </feedback_content>
</feedback>

Ltac-debug messages

Ltac-debug messages are issued out-of-band, similar to Feedback messages. The response contains an identifying tag and a <ppdoc>. Currently these tags are used:

  • output - ordinary output for display in the Messages panel
  • goal - the current goal for the debugger, for display in the Messages panel or elsewhere
  • prompt - output for display in the Messages panel prompting the user to enter a debug command, allowing CoqIDE to display it without appending a newline. It also signals that coqidetop is waiting to receive a debugger-specific message such as Db_cmd.
<ltac_debug>
prompt
  <ppdoc val="tag">
        :
  </ppdoc>
</ltac_debug>

Highlighting Text

Proof diffs highlight differences between the current and previous proof states in the displayed output. These are represented by tags embedded in output fields of the XML document.

There are 4 tags that indicate how the enclosed text should be highlighted:

  • diff.added - added text
  • diff.removed - removed text
  • diff.added.bg - unchanged text in a line that has additions ("bg" for "background")
  • diff.removed.bg - unchanged text in a line that has removals

CoqIDE, Proof General and coqtop currently use 2 shades of green and 2 shades of red as the background color for highlights. Coqtop and CoqIDE also apply underlining and/or strikeout highlighting for the sake of the color blind.

For example, <diff.added>ABC</diff.added> indicates that "ABC" should be highlighted as added text. Tags can be nested, such as: <diff.added.bg>A <diff.added> + 1</diff.added> + B</diff.added.bg>. IDE code displaying highlighted strings should maintain a stack for nested tags and the associated highlight. Currently the diff code only nests at most 2 tags deep. If an IDE uses other highlights such as text foreground color or italic text, it may need to merge the background color with those other highlights to give the desired (IDE dependent) behavior.

The current implementations avoid highlighting white space at the beginning or the end of a line. This gives a better appearance.

There may be additional text that is marked with other tags in the output text. IDEs probably should ignore and not display tags they don't recognize.

Some internal details about generating tags within Coq (e.g. if you want to add additional tags):

Tagged output strings are generated from Pp.t's. Use Pp.tag to highlight a Pp.t using one of the tags listed above. A span of tokens can be marked by using "start." on the first token and "end." on the last token. (Span markers are needed because a span of tokens in the output may not match nesting of layout boxes in the Pp.t.) The conversion from the Pp.t to the XML-tagged string replaces the "start.*" and "end.*" tags with the basic tags.