/*
* Jalview - A Sequence Alignment Editor and Viewer (Version 2.9)
* Copyright (C) 2015 The Jalview Authors
*
* This file is part of Jalview.
*
* Jalview is free software: you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation, either version 3
* of the License, or (at your option) any later version.
*
* Jalview is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty
* of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with Jalview. If not, see .
* The Jalview Authors are detailed in the 'AUTHORS' file.
*/
package jalview.ws.rest;
import jalview.bin.Cache;
import jalview.datamodel.Alignment;
import jalview.datamodel.AlignmentAnnotation;
import jalview.datamodel.AlignmentI;
import jalview.datamodel.AlignmentOrder;
import jalview.datamodel.Annotation;
import jalview.datamodel.ColumnSelection;
import jalview.datamodel.SequenceGroup;
import jalview.datamodel.SequenceI;
import jalview.gui.AlignFrame;
import jalview.gui.Desktop;
import jalview.gui.PaintRefresher;
import jalview.gui.WebserviceInfo;
import jalview.io.NewickFile;
import jalview.io.packed.JalviewDataset;
import jalview.io.packed.JalviewDataset.AlignmentSet;
import jalview.util.MessageManager;
import jalview.ws.AWSThread;
import jalview.ws.AWsJob;
import java.awt.event.ActionEvent;
import java.awt.event.ActionListener;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Hashtable;
import java.util.List;
import java.util.Map.Entry;
import org.apache.axis.transport.http.HTTPConstants;
import org.apache.http.Header;
import org.apache.http.HttpEntity;
import org.apache.http.HttpResponse;
import org.apache.http.client.ClientProtocolException;
import org.apache.http.client.methods.HttpGet;
import org.apache.http.client.methods.HttpPost;
import org.apache.http.client.methods.HttpRequestBase;
import org.apache.http.entity.mime.HttpMultipartMode;
import org.apache.http.entity.mime.MultipartEntity;
import org.apache.http.impl.client.DefaultHttpClient;
import org.apache.http.protocol.BasicHttpContext;
import org.apache.http.protocol.HttpContext;
import org.apache.http.util.EntityUtils;
public class RestJobThread extends AWSThread
{
enum Stage
{
SUBMIT, POLL
}
protected RestClient restClient;
public RestJobThread(RestClient restClient)
{
super(restClient.af, null, restClient._input,
restClient.service.postUrl);
this.restClient = restClient; // may not be needed
// Test Code
// minimal job - submit given input and parse result onto alignment as
// annotation/whatever
// look for tree data, etc.
// for moment do following job type only
// input=visiblealignment,groupsindex
// ie one subjob using groups defined on alignment.
if (!restClient.service.isHseparable())
{
jobs = new RestJob[1];
jobs[0] = new RestJob(0, this,
restClient._input.getVisibleAlignment(restClient.service
.getGapCharacter()),
restClient._input.getVisibleContigs());
// need a function to get a range on a view/alignment and return both
// annotation, groups and selection subsetted to just that region.
}
else
{
int[] viscontig = restClient._input.getVisibleContigs();
AlignmentI[] viscontigals = restClient._input
.getVisibleContigAlignments(restClient.service
.getGapCharacter());
if (viscontigals != null && viscontigals.length > 0)
{
jobs = new RestJob[viscontigals.length];
for (int j = 0; j < jobs.length; j++)
{
int[] visc = new int[] { viscontig[j * 2], viscontig[j * 2 + 1] };
if (j != 0)
{
jobs[j] = new RestJob(j, this, viscontigals[j], visc);
}
else
{
jobs[j] = new RestJob(0, this, viscontigals[j], visc);
}
}
}
}
// end Test Code
/**
* subjob types row based: per sequence in alignment/selected region { input
* is one sequence or sequence ID } per alignment/selected region { input is
* set of sequences, alignment, one or more sets of sequence IDs,
*/
if (!restClient.service.isHseparable())
{
// create a bunch of subjobs per visible contig to ensure result honours
// hidden boundaries
// TODO: determine if a 'soft' hSeperable property is needed - e.g. if
// user does SS-pred on sequence with big hidden regions, its going to be
// less reliable.
}
else
{
// create a single subjob for the visible/selected input
}
// TODO: decide if vSeperable exists: eg. column wide analysis where hidden
// rows do not affect output - generally no analysis that generates
// alignment annotation is vSeparable -
}
/**
* create gui components for monitoring jobs
*
* @param webserviceInfo
*/
public void setWebServiceInfo(WebserviceInfo webserviceInfo)
{
wsInfo = webserviceInfo;
for (int j = 0; j < jobs.length; j++)
{
wsInfo.addJobPane();
// Copy over any per job params
if (jobs.length > 1)
{
wsInfo.setProgressName("region " + jobs[j].getJobnum(),
jobs[j].getJobnum());
}
else
{
wsInfo.setProgressText(jobs[j].getJobnum(), OutputHeader);
}
}
}
private String getStage(Stage stg)
{
if (stg == Stage.SUBMIT)
return "submitting ";
if (stg == Stage.POLL)
return "checking status of ";
return (" being confused about ");
}
private void doPoll(RestJob rj) throws Exception
{
String postUrl = rj.getPollUrl();
doHttpReq(Stage.POLL, rj, postUrl);
}
/**
* construct the post and handle the response.
*
* @throws Exception
*/
public void doPost(RestJob rj) throws Exception
{
String postUrl = rj.getPostUrl();
doHttpReq(Stage.SUBMIT, rj, postUrl);
wsInfo.invalidate();
}
/**
* do the HTTP request - and respond/set statuses appropriate to the current
* stage.
*
* @param stg
* @param rj
* - provides any data needed for posting and used to record state
* @param postUrl
* - actual URL to post/get from
* @throws Exception
*/
protected void doHttpReq(Stage stg, RestJob rj, String postUrl)
throws Exception
{
StringBuffer respText = new StringBuffer();
// con.setContentHandlerFactory(new
// jalview.ws.io.mime.HttpContentHandler());
HttpRequestBase request = null;
String messages = "";
if (stg == Stage.SUBMIT)
{
// Got this from
// http://evgenyg.wordpress.com/2010/05/01/uploading-files-multipart-post-apache/
HttpPost htpost = new HttpPost(postUrl);
MultipartEntity postentity = new MultipartEntity(
HttpMultipartMode.STRICT);
for (Entry input : rj.getInputParams())
{
if (input.getValue().validFor(rj))
{
postentity.addPart(input.getKey(), input.getValue()
.formatForInput(rj));
}
else
{
messages += "Skipped an input (" + input.getKey()
+ ") - Couldn't generate it from available data.";
}
}
htpost.setEntity(postentity);
request = htpost;
}
else
{
request = new HttpGet(postUrl);
}
if (request != null)
{
DefaultHttpClient httpclient = new DefaultHttpClient();
HttpContext localContext = new BasicHttpContext();
HttpResponse response = null;
try
{
response = httpclient.execute(request);
} catch (ClientProtocolException he)
{
rj.statMessage = "Web Protocol Exception when " + getStage(stg)
+ "Job.
Problematic url was " + request.getURI()
+ "
See Console output for details.";
rj.setAllowedServerExceptions(0);// unrecoverable;
rj.error = true;
Cache.log.fatal("Unexpected REST Job " + getStage(stg)
+ "exception for URL " + rj.rsd.postUrl);
throw (he);
} catch (IOException e)
{
rj.statMessage = "IO Exception when " + getStage(stg)
+ "Job.
Problematic url was " + request.getURI()
+ "
See Console output for details.";
Cache.log.warn("IO Exception for REST Job " + getStage(stg)
+ "exception for URL " + rj.rsd.postUrl);
throw (e);
}
switch (response.getStatusLine().getStatusCode())
{
case 200:
rj.running = false;
Cache.log.debug("Processing result set.");
processResultSet(rj, response, request);
break;
case 202:
rj.statMessage = "
Job submitted successfully. Results available at this URL:\n"
+ ""
+ rj.getJobId()
+ "
";
rj.running = true;
break;
case 302:
Header[] loc;
if (!rj.isSubmitted()
&& (loc = response
.getHeaders(HTTPConstants.HEADER_LOCATION)) != null
&& loc.length > 0)
{
if (loc.length > 1)
{
Cache.log
.warn("Ignoring additional "
+ (loc.length - 1)
+ " location(s) provided in response header ( next one is '"
+ loc[1].getValue() + "' )");
}
rj.setJobId(loc[0].getValue());
rj.setSubmitted(true);
}
completeStatus(rj, response);
break;
case 500:
// Failed.
rj.setSubmitted(true);
rj.setAllowedServerExceptions(0);
rj.setSubjobComplete(true);
rj.error = true;
rj.running = false;
completeStatus(rj, response, "" + getStage(stg)
+ "failed. Reason below:\n");
break;
default:
// Some other response. Probably need to pop up the content in a window.
// TODO: deal with all other HTTP response codes from server.
Cache.log.warn("Unhandled response status when " + getStage(stg)
+ "for " + postUrl + ": " + response.getStatusLine());
rj.error = true;
rj.setAllowedServerExceptions(0);
rj.setSubjobComplete(true);
rj.setSubmitted(true);
try
{
completeStatus(
rj,
response,
""
+ getStage(stg)
+ " resulted in an unexpected server response.
Url concerned was "
+ request.getURI()
+ "
Filtered response content below:
");
} catch (IOException e)
{
Cache.log.debug("IOException when consuming unhandled response",
e);
}
;
}
}
}
/**
* job has completed. Something valid should be available from con
*
* @param rj
* @param con
* @param req
* is a stateless request - expected to return the same data
* regardless of how many times its called.
*/
private void processResultSet(RestJob rj, HttpResponse con,
HttpRequestBase req)
{
if (rj.statMessage == null)
{
rj.statMessage = "";
}
rj.statMessage += "Job Complete.\n";
try
{
rj.resSet = new HttpResultSet(rj, con, req);
rj.gotresult = true;
} catch (IOException e)
{
rj.statMessage += "Couldn't parse results. Failed.";
rj.error = true;
rj.gotresult = false;
}
}
private void completeStatus(RestJob rj, HttpResponse con)
throws IOException
{
completeStatus(rj, con, null);
}
private void completeStatus(RestJob rj, HttpResponse con, String prefix)
throws IOException
{
StringBuffer sb = new StringBuffer();
if (prefix != null)
{
sb.append(prefix);
}
;
if (rj.statMessage != null && rj.statMessage.length() > 0)
{
sb.append(rj.statMessage);
}
HttpEntity en = con.getEntity();
/*
* Just append the content as a string.
*/
String f;
StringBuffer content = new StringBuffer(f = EntityUtils.toString(en));
f = f.toLowerCase();
int body = f.indexOf(" -1)
{
content.delete(0, f.indexOf(">", body) + 1);
}
if (body > -1 && sb.length() > 0)
{
sb.append("\n");
content.insert(0, sb);
sb = null;
}
f = null;
rj.statMessage = content.toString();
}
@Override
public void pollJob(AWsJob job) throws Exception
{
assert (job instanceof RestJob);
System.err.println("Debug RestJob: Polling Job");
doPoll((RestJob) job);
}
@Override
public void StartJob(AWsJob job)
{
assert (job instanceof RestJob);
try
{
System.err.println("Debug RestJob: Posting Job");
doPost((RestJob) job);
} catch (NoValidInputDataException erex)
{
job.setSubjobComplete(true);
job.setSubmitted(true);
((RestJob) job).statMessage = "
It looks like there was a problem with the data sent to the service :
"
+ erex.getMessage() + "\n";
((RestJob) job).error = true;
} catch (Exception ex)
{
job.setSubjobComplete(true);
job.setAllowedServerExceptions(-1);
Cache.log.error("Exception when trying to start Rest Job.", ex);
}
}
@Override
public void parseResult()
{
// crazy users will see this message
// TODO: finish this! and remove the message below!
Cache.log.warn("Rest job result parser is currently INCOMPLETE!");
int validres = 0;
for (RestJob rj : (RestJob[]) jobs)
{
if (rj.hasResponse() && rj.resSet != null && rj.resSet.isValid())
{
String ln = null;
try
{
Cache.log.debug("Parsing data for job " + rj.getJobId());
rj.parseResultSet();
if (rj.hasResults())
{
validres++;
}
Cache.log.debug("Finished parsing data for job " + rj.getJobId());
} catch (Error ex)
{
Cache.log.warn("Failed to finish parsing data for job "
+ rj.getJobId());
ex.printStackTrace();
} catch (Exception ex)
{
Cache.log.warn("Failed to finish parsing data for job "
+ rj.getJobId());
ex.printStackTrace();
} finally
{
rj.error = true;
rj.statMessage = "Error whilst parsing data for this job.
URL for job response is :"
+ rj.resSet.getUrl()
+ "
";
}
}
}
if (validres > 0)
{
// add listeners and activate result display gui elements
/**
* decisions based on job result content + state of alignFrame that
* originated the job:
*/
/*
* 1. Can/Should this job automatically open a new window for results
*/
if (true)
{
// preserver current jalview behaviour
wsInfo.setViewResultsImmediatly(true);
}
else
{
// realiseResults(true, true);
}
// otherwise, should automatically view results
// TODO: check if at least one or more contexts are valid - if so, enable
// gui
wsInfo.showResultsNewFrame.addActionListener(new ActionListener()
{
@Override
public void actionPerformed(ActionEvent e)
{
realiseResults(false);
}
});
wsInfo.mergeResults.addActionListener(new ActionListener()
{
@Override
public void actionPerformed(ActionEvent e)
{
realiseResults(true);
}
});
wsInfo.setResultsReady();
}
else
{
// tell the user nothing was returned.
wsInfo.setStatus(wsInfo.STATE_STOPPED_ERROR);
wsInfo.setFinishedNoResults();
}
}
/**
* instructions for whether to create new alignment view on current alignment
* set, add to current set, or create new alignFrame
*/
private enum AddDataTo
{
/**
* add annotation, trees etc to current view
*/
currentView,
/**
* create a new view derived from current view and add data to that
*/
newView,
/**
* create a new alignment frame for the result set and add annotation to
* that.
*/
newAlignment
};
public void realiseResults(boolean merge)
{
/*
* 2. Should the job modify the parent alignment frame/view(s) (if they
* still exist and the alignment hasn't been edited) in order to display new
* annotation/features.
*/
/**
* alignment panels derived from each alignment set returned by service.
*/
ArrayList destPanels = new ArrayList();
/**
* list of instructions for how to process each distinct alignment set
* returned by the job set
*/
ArrayList resultDest = new ArrayList();
/**
* when false, zeroth pane is panel derived from input deta.
*/
boolean newAlignment = false;
/**
* gap character to be used for alignment reconstruction
*/
char gapCharacter = restClient.av.getGapCharacter();
// Now, iterate over all alignment sets returned from all jobs:
// first inspect jobs and collate results data in order to count alignments
// and other results
// then assemble results from decomposed (v followed by h-separated) jobs
// finally, new views and alignments will be created and displayed as
// necessary.
boolean hsepjobs = restClient.service.isHseparable();
boolean vsepjobs = restClient.service.isVseparable();
// total number of distinct alignment sets generated by job set.
int numAlSets = 0, als = 0;
List destAls = new ArrayList();
List destColsel = new ArrayList();
List> trees = new ArrayList>();
do
{
// Step 1.
// iterate over each alignment set returned from each subjob. Treating
// each one returned in parallel with others.
// Result collation arrays
/**
* mapping between index of sequence in alignment that was submitted to
* server and index of sequence in the input alignment
*/
int[][] ordermap = new int[jobs.length][];
SequenceI[][] rseqs = new SequenceI[jobs.length][];
AlignmentOrder[] orders = new AlignmentOrder[jobs.length];
AlignmentAnnotation[][] alan = new AlignmentAnnotation[jobs.length][];
SequenceGroup[][] sgrp = new SequenceGroup[jobs.length][];
// Now collect all data for alignment Set als from job array
for (int j = 0; j < jobs.length; j++)
{
RestJob rj = (RestJob) jobs[j];
if (rj.hasResults())
{
JalviewDataset rset = rj.context;
if (rset.hasAlignments())
{
if (numAlSets < rset.getAl().size())
{
numAlSets = rset.getAl().size();
}
if (als < rset.getAl().size()
&& rset.getAl().get(als).isModified())
{
// Collate result data
// TODO: decide if all alignmentI should be collected rather than
// specific alignment data containers
// for moment, we just extract content, but this means any
// alignment properties may be lost.
AlignmentSet alset = rset.getAl().get(als);
alan[j] = alset.al.getAlignmentAnnotation();
if (alset.al.getGroups() != null)
{
sgrp[j] = new SequenceGroup[alset.al.getGroups().size()];
alset.al.getGroups().toArray(sgrp[j]);
}
else
{
sgrp[j] = null;
}
orders[j] = new AlignmentOrder(alset.al);
rseqs[j] = alset.al.getSequencesArray();
ordermap[j] = rj.getOrderMap();
// if (rj.isInputUniquified()) {
// jalview.analysis.AlignmentSorter.recoverOrder(rseqs[als]);
// }
if (alset.trees != null)
{
trees.add(new ArrayList(alset.trees));
}
else
{
trees.add(new ArrayList());
}
}
}
}
}
// Now aggregate and present results from this frame of alignment data.
int nvertsep = 0, nvertseps = 1;
if (vsepjobs)
{
// Jobs relate to different rows of input alignment.
// Jobs are subdivided by rows before columns,
// so there will now be a number of subjobs according tohsep for each
// vertsep
// TODO: get vertical separation intervals for each job and set
// nvertseps
// TODO: merge data from each group/sequence onto whole
// alignment
}
/**
* index into rest jobs subdivided vertically
*/
int vrestjob = 0;
// Destination alignments for all result data.
ArrayList visgrps = new ArrayList();
Hashtable groupNames = new Hashtable();
ArrayList visAlAn = null;
for (nvertsep = 0; nvertsep < nvertseps; nvertsep++)
{
// TODO: set scope w.r.t. original alignment view for vertical
// separation.
{
// results for a job exclude hidden columns of input data, so map
// back on to all visible regions
/**
* rest job result we are working with
*/
int nrj = vrestjob;
RestJob rj = (RestJob) jobs[nrj];
int contigs[] = input.getVisibleContigs();
AlignmentI destAl = null;
jalview.datamodel.ColumnSelection destCs = null;
// Resolve destAl for this data.
if (als == 0 && rj.isInputContextModified())
{
// special case: transfer features, annotation, groups, etc,
// from input
// context to align panel derived from input data
if (destAls.size() > als)
{
destAl = destAls.get(als);
}
else
{
if (!restClient.isAlignmentModified() && merge)
{
destAl = restClient.av.getAlignment();
destCs = restClient.av.getColumnSelection();
resultDest
.add(restClient.isShowResultsInNewView() ? AddDataTo.newView
: AddDataTo.currentView);
destPanels.add(restClient.recoverAlignPanelForView());
}
else
{
newAlignment = true;
// recreate the input alignment data
Object[] idat = input
.getAlignmentAndColumnSelection(gapCharacter);
destAl = new Alignment((SequenceI[]) idat[0]);
destCs = (ColumnSelection) idat[1];
resultDest.add(AddDataTo.newAlignment);
// but do not add to the alignment panel list - since we need to
// create a whole new alignFrame set.
}
destAls.add(destAl);
destColsel.add(destCs);
}
}
else
{
// alignment(s) returned by service is to be re-integrated and
// displayed
if (destAls.size() > als)
{
if (!vsepjobs)
{
// TODO: decide if multiple multiple alignments returned by
// non-vseparable services are allowed.
Cache.log
.warn("dealing with multiple alignment products returned by non-vertically separable service.");
}
// recover reference to last alignment created for this rest frame
// ready for extension
destAl = destAls.get(als);
destCs = destColsel.get(als);
}
else
{
Object[] newview;
if (!hsepjobs)
{
// single alignment for any job that gets mapped back on to
// input data. Reconstruct by interleaving parts of returned
// alignment with hidden parts of input data.
SequenceI[][] nsq = splitSeqsOnVisibleContigs(rseqs[nrj],
contigs, gapCharacter);
AlignmentOrder alo[] = new AlignmentOrder[nsq.length];
for (int no = 0; no < alo.length; no++)
{
alo[no] = new AlignmentOrder(orders[nrj].getOrder());
}
newview = input.getUpdatedView(nsq, orders, gapCharacter);
}
else
{
// each job maps to a single visible contig, and all need to be
// stitched back together.
// reconstruct using sub-region based MSA alignment construction
// mechanism
newview = input.getUpdatedView(rseqs, orders, gapCharacter);
}
destAl = new Alignment((SequenceI[]) newview[0]);
destCs = (ColumnSelection) newview[1];
newAlignment = true;
// TODO create alignment from result data with propagated
// references.
destAls.add(destAl);
destColsel.add(destCs);
resultDest.add(AddDataTo.newAlignment);
throw new Error(
MessageManager
.getString("error.implementation_error")
+ "TODO: ");
}
}
/**
* save initial job in this set in case alignment is h-separable
*/
int initnrj = nrj;
// Now add in groups
for (int ncnt = 0; ncnt < contigs.length; ncnt += 2)
{
if (!hsepjobs)
{
// single alignment for any job that gets mapped back on to input
// data.
}
else
{
// each job maps to a single visible contig, and all need to be
// stitched back together.
if (ncnt > 0)
{
nrj++;
}
// TODO: apply options for group merging and annotation merging.
// If merging not supported, then either clear hashtables now or
// use them to rename the new annotation/groups for each contig if
// a conflict occurs.
}
if (sgrp[nrj] != null)
{
for (SequenceGroup sg : sgrp[nrj])
{
boolean recovered = false;
SequenceGroup exsg = null;
if (sg.getName() != null)
{
exsg = groupNames.get(sg.getName());
}
if (exsg == null)
{
exsg = new SequenceGroup(sg);
groupNames.put(exsg.getName(), exsg);
visgrps.add(exsg);
exsg.setStartRes(sg.getStartRes() + contigs[ncnt]);
exsg.setEndRes(sg.getEndRes() + contigs[ncnt]);
}
else
{
recovered = true;
}
// now replace any references from the result set with
// corresponding refs from alignment input set.
// TODO: cope with recovering hidden sequences from
// resultContext
{
for (SequenceI oseq : sg.getSequences(null))
{
SequenceI nseq = getNewSeq(oseq, rseqs[nrj],
ordermap[nrj], destAl);
if (nseq != null)
{
if (!recovered)
{
exsg.deleteSequence(oseq, false);
}
exsg.addSequence(nseq, false);
}
else
{
Cache.log
.warn("Couldn't resolve original sequence for new sequence.");
}
}
if (sg.hasSeqrep())
{
if (exsg.getSeqrep() == sg.getSeqrep())
{
// lift over sequence rep reference too
SequenceI oseq = sg.getSeqrep();
SequenceI nseq = getNewSeq(oseq, rseqs[nrj],
ordermap[nrj], destAl);
if (nseq != null)
{
exsg.setSeqrep(nseq);
}
}
}
}
if (recovered)
{
// adjust boundaries of recovered group w.r.t. new group being
// merged on to original alignment.
int start = sg.getStartRes() + contigs[ncnt], end = sg
.getEndRes() + contigs[ncnt];
if (start < exsg.getStartRes())
{
exsg.setStartRes(start);
}
if (end > exsg.getEndRes())
{
exsg.setEndRes(end);
}
}
}
}
}
// reset job counter
nrj = initnrj;
// and finally add in annotation and any trees for each job
for (int ncnt = 0; ncnt < contigs.length; ncnt += 2)
{
if (!hsepjobs)
{
// single alignment for any job that gets mapped back on to input
// data.
}
else
{
// each job maps to a single visible contig, and all need to be
// stitched back together.
if (ncnt > 0)
{
nrj++;
}
}
// merge alignmentAnnotation into one row
if (alan[nrj] != null)
{
for (int an = 0; an < alan[nrj].length; an++)
{
SequenceI sqass = null;
SequenceGroup grass = null;
if (alan[nrj][an].sequenceRef != null)
{
// TODO: ensure this relocates sequence reference to local
// context.
sqass = getNewSeq(alan[nrj][an].sequenceRef, rseqs[nrj],
ordermap[nrj], destAl);
}
if (alan[nrj][an].groupRef != null)
{
// TODO: verify relocate group reference to local context
// works correctly
grass = groupNames.get(alan[nrj][an].groupRef.getName());
if (grass == null)
{
Cache.log
.error("Couldn't relocate group referemce for group "
+ alan[nrj][an].groupRef.getName());
}
}
if (visAlAn == null)
{
visAlAn = new ArrayList();
}
AlignmentAnnotation visan = null;
for (AlignmentAnnotation v : visAlAn)
{
if (v.label != null
&& v.label.equals(alan[nrj][an].label))
{
visan = v;
}
}
if (visan == null)
{
visan = new AlignmentAnnotation(alan[nrj][an]);
// copy annotations, and wipe out/update refs.
visan.annotations = new Annotation[input.getWidth()];
visan.groupRef = grass;
visan.sequenceRef = sqass;
visAlAn.add(visan);
}
if (contigs[ncnt] + alan[nrj][an].annotations.length > visan.annotations.length)
{
// increase width of annotation row
Annotation[] newannv = new Annotation[contigs[ncnt]
+ alan[nrj][an].annotations.length];
System.arraycopy(visan.annotations, 0, newannv, 0,
visan.annotations.length);
visan.annotations = newannv;
}
// now copy local annotation data into correct position
System.arraycopy(alan[nrj][an].annotations, 0,
visan.annotations, contigs[ncnt],
alan[nrj][an].annotations.length);
}
}
// Trees
if (trees.get(nrj).size() > 0)
{
for (NewickFile nf : trees.get(nrj))
{
// TODO: process each newick file, lifting over sequence refs to
// current alignment, if necessary.
Cache.log
.error("Tree recovery from restjob not yet implemented.");
}
}
}
}
} // end of vseps loops.
if (visAlAn != null)
{
for (AlignmentAnnotation v : visAlAn)
{
destAls.get(als).addAnnotation(v);
}
}
if (visgrps != null)
{
for (SequenceGroup sg : visgrps)
{
destAls.get(als).addGroup(sg);
}
}
} while (++als < numAlSets);
// Finally, assemble each new alignment, and create new gui components to
// present it.
/**
* current AlignFrame where results will go.
*/
AlignFrame destaf = restClient.recoverAlignFrameForView();
/**
* current pane being worked with
*/
jalview.gui.AlignmentPanel destPanel = restClient
.recoverAlignPanelForView();
als = 0;
for (AddDataTo action : resultDest)
{
AlignmentI destal;
ColumnSelection destcs;
String alTitle = MessageManager.formatMessage(
"label.webservice_job_title_on", new String[] {
restClient.service.details.Action,
restClient.service.details.Name, restClient.viewTitle });
switch (action)
{
case newAlignment:
destal = destAls.get(als);
destcs = destColsel.get(als);
destaf = new AlignFrame(destal, destcs, AlignFrame.DEFAULT_WIDTH,
AlignFrame.DEFAULT_HEIGHT);
PaintRefresher.Refresh(destaf, destaf.getViewport()
.getSequenceSetId());
// todo transfer any feature settings and colouring
/*
* destaf.getFeatureRenderer().transferSettings(this.featureSettings);
* // update orders if (alorders.size() > 0) { if (alorders.size() == 1)
* { af.addSortByOrderMenuItem(WebServiceName + " Ordering",
* (AlignmentOrder) alorders.get(0)); } else { // construct a
* non-redundant ordering set Vector names = new Vector(); for (int i =
* 0, l = alorders.size(); i < l; i++) { String orderName = new
* String(" Region " + i); int j = i + 1;
*
* while (j < l) { if (((AlignmentOrder) alorders.get(i))
* .equals(((AlignmentOrder) alorders.get(j)))) { alorders.remove(j);
* l--; orderName += "," + j; } else { j++; } }
*
* if (i == 0 && j == 1) { names.add(new String("")); } else {
* names.add(orderName); } } for (int i = 0, l = alorders.size(); i < l;
* i++) { af.addSortByOrderMenuItem( WebServiceName + ((String)
* names.get(i)) + " Ordering", (AlignmentOrder) alorders.get(i)); } } }
*/
// TODO: modify this and previous alignment's title if many alignments
// have been returned.
Desktop.addInternalFrame(destaf, alTitle, AlignFrame.DEFAULT_WIDTH,
AlignFrame.DEFAULT_HEIGHT);
break;
case newView:
// TODO: determine title for view
break;
case currentView:
break;
}
}
if (!newAlignment)
{
if (restClient.isShowResultsInNewView())
{
// destPanel = destPanel.alignFrame.newView(false);
}
}
else
{
}
/*
* if (als) // add the destination panel to frame zero of result panel set }
* } if (destPanels.size()==0) { AlignFrame af = new AlignFrame((AlignmentI)
* idat[0], (ColumnSelection) idat[1], AlignFrame.DEFAULT_WIDTH,
* AlignFrame.DEFAULT_HEIGHT);
*
* jalview.gui.Desktop.addInternalFrame(af, "Results for " +
* restClient.service.details.Name + " " + restClient.service.details.Action
* + " on " + restClient.af.getTitle(), AlignFrame.DEFAULT_WIDTH,
* AlignFrame.DEFAULT_HEIGHT); destPanel = af.alignPanel; // create totally
* new alignment from stashed data/results
*/
/*
*/
/**
* alignments. New alignments are added to dataset, and subsequently
* annotated/visualised accordingly. 1. New alignment frame created for new
* alignment. Decide if any vis settings should be inherited from old
* alignment frame (e.g. sequence ordering ?). 2. Subsequent data added to
* alignment as below:
*/
/**
* annotation update to original/newly created context alignment: 1.
* identify alignment where annotation is to be loaded onto. 2. Add
* annotation, excluding any duplicates. 3. Ensure annotation is visible on
* alignment - honouring ordering given by file.
*/
/**
* features updated to original or newly created context alignment: 1.
* Features are(or were already) added to dataset. 2. Feature settings
* modified to ensure all features are displayed - honouring any ordering
* given by result file. Consider merging action with the code used by the
* DAS fetcher to update alignment views with new info.
*/
/**
* Seq associated data files (PDB files). 1. locate seq association in
* current dataset/alignment context and add file as normal - keep handle of
* any created ref objects. 2. decide if new data should be displayed : PDB
* display: if alignment has PDB display already, should new pdb files be
* aligned to it ?
*
*/
// destPanel.adjustAnnotationHeight();
}
/**
* split the given array of sequences into blocks of subsequences
* corresponding to each visible contig
*
* @param sequenceIs
* @param contigs
* @param gapChar
* padding character for ragged ends of visible contig region.
* @return
*/
private SequenceI[][] splitSeqsOnVisibleContigs(SequenceI[] sequenceIs,
int[] contigs, char gapChar)
{
int nvc = contigs == null ? 1 : contigs.length / 2;
SequenceI[][] blocks = new SequenceI[nvc][];
if (contigs == null)
{
blocks[0] = new SequenceI[sequenceIs.length];
System.arraycopy(sequenceIs, 0, blocks[0], 0, sequenceIs.length);
}
else
{
// deja vu - have I written this before ?? propagateGaps does this in a
// way
char[] gapset = null;
int start = 0, width = 0;
for (int c = 0; c < nvc; c++)
{
width = contigs[c * 2 + 1] - contigs[c * 2] + 1;
for (int s = 0; s < sequenceIs.length; s++)
{
int end = sequenceIs[s].getLength();
if (start < end)
{
if (start + width < end)
{
blocks[c][s] = sequenceIs[s].getSubSequence(start, start
+ width);
}
else
{
blocks[c][s] = sequenceIs[s].getSubSequence(start, end);
String sq = blocks[c][s].getSequenceAsString();
for (int n = start + width; n > end; n--)
{
sq += gapChar;
}
}
}
else
{
if (gapset == null || gapset.length < width)
{
char ng[] = new char[width];
int gs = 0;
if (gapset != null)
{
System.arraycopy(gapset, 0, ng, 0, gs = gapset.length);
}
while (gs < ng.length)
{
ng[gs++] = gapChar;
}
}
blocks[c][s] = sequenceIs[s].getSubSequence(end, end);
blocks[c][s].setSequence(gapset.toString().substring(0, width));
}
}
if (c > 0)
{
// adjust window for next visible segnment
start += contigs[c * 2 + 1] - contigs[c * 2];
}
}
}
return blocks;
}
/**
* recover corresponding sequence from original input data corresponding to
* sequence in a specific job's input data.
*
* @param oseq
* @param sequenceIs
* @param is
* @param destAl
* @return
*/
private SequenceI getNewSeq(SequenceI oseq, SequenceI[] sequenceIs,
int[] is, AlignmentI destAl)
{
int p = 0;
while (p < sequenceIs.length && sequenceIs[p] != oseq)
{
p++;
}
if (p < sequenceIs.length && p < destAl.getHeight())
{
return destAl.getSequenceAt(is[p]);
}
return null;
}
/**
*
* @return true if the run method is safe to call
*/
public boolean isValid()
{
ArrayList _warnings = new ArrayList();
boolean validt = true;
if (jobs != null)
{
for (RestJob rj : (RestJob[]) jobs)
{
if (!rj.hasValidInput())
{
// invalid input for this job
System.err.println("Job " + rj.getJobnum()
+ " has invalid input. ( " + rj.getStatus() + ")");
if (rj.hasStatus() && !_warnings.contains(rj.getStatus()))
{
_warnings.add(rj.getStatus());
}
validt = false;
}
}
}
if (!validt)
{
warnings = "";
for (String st : _warnings)
{
if (warnings.length() > 0)
{
warnings += "\n";
}
warnings += st;
}
}
return validt;
}
protected String warnings;
public boolean hasWarnings()
{
// TODO Auto-generated method stub
return warnings != null && warnings.length() > 0;
}
/**
* get any informative messages about why the job thread couldn't start.
*
* @return
*/
public String getWarnings()
{
return isValid() ? "Job can be started. No warnings."
: hasWarnings() ? warnings : "";
}
}