broken node name when plot gating tree on mac

[mikejiang]

The node name (when it has minus sign) in the graph is corrupted when graphviz library tries to convert dot file to gxl file.

  DotFile<-tempfile(fileext=".dot")
  .Call("R_plotGh",x@pointer,getSample(x),DotFile,FALSE)
  GXLFile<-tempfile(fileext=".gxl")
  system(paste("dot2gxl",DotFile, ">>",GXLFile))

It is mac-specific issue.

[mikejiang]

When dot2gxl command converts the same graph file from dot to gxl format, it keeps the node name as it is on linux ( e.g. v-) , but encoded it to v- on mac. Then somehow graph::fromGXL doesn't recognize the encoded node name. So it is the gxl parser from graph that corrupted the name. But the platform-specific behaviour of graphviz converter also confounded this issue.

[gfinak]

Anything we can do to get this working properly?

On Wed, Nov 20, 2013 at 12:02 PM, Mike Jiang notifications@github.comwrote:

When dot2gxl command converts the same graph file from dot to gxl format, it keeps the node name as it is on linux ( e.g. v-) , but encoded it to v- on mac. Then somehow graph::fromGXL doesn't recognize the encoded node name. So it is the gxl parser from graph that corrupted the name. But the platform-specific behaviour of graphviz converter also confound this issue.

— Reply to this email directly or view it on GitHubhttps://github.com/RGLab/flowWorkspace/issues/114#issuecomment-28925016 .

Greg Finak, PhD Staff Scientist Vaccine and Infectious Disease Division Fred Hutchinson Cancer Research Center Seattle, WA 98109

(206)667-3116 gfinak@fhcrc.org

[mikejiang]

I am looking into fromGXL call to see if there is an easy fix.

[mikejiang]

graph::fromGXL uses XML:xmlEventParse to parse each tag from gxl (as shown below, essentially an xml file)

<gxl>
        <graph id="G" edgeids="true" edgemode="directed">
                <node id="N_0">
                        <attr name="label">
                                <string>root</string>
                        </attr>
                </node>
                <node id="N_1">
                        <attr name="label">
                                <string>&#45;v</string>
                        </attr>
                </node>
        </graph>
</gxl>

Here is what happened based on the diagnostic output of graph:::graph_handler: Whenever xmlEventParse encounters the node name (stored in /gxl/graph/node/attr/string tag) that contains the numeric character reference (i.e. unicode &#nnnn;), it

• decode it into ASCII
• and splits the content into string array instead of concatenated them into single string (source of evil)

when graph:::graph_handler tries to collect the node name from the parsed tag content,

• it assumes there is only one string per tag
• thus mistakenly reads split character vector multiple times
• with the last character overwriting all the previous ones.
• then we ended up seeing partial (-) instead of complete node name (v-)

The quick fix will be hacking into graph:::graph_handler to concatenate these characters when they are coming from the same attribute of the same node.

[kevinushey]

Platform specific issues are always the best...

Thanks for investigating this!

On Wed, Nov 20, 2013 at 4:05 PM, Mike Jiang notifications@github.comwrote:

graph::fromGXL uses XML:xmlEventParse to parse each tag from gxl (as shown below, essentially an xml file)

root -v

Here is what happened based on the diagnostic output of graph:::graph_handler: Whenever xmlEventParse encounters the node name (stored in /gxl/graph/node/attr/string tag) that contains the numeric character reference (i.e. unicode &#nnnn;), it

• decode it into ASCII
• and splits the content into string array instead of concatenated them into single string (source of evil)

when graph:::graph_handler tries to collect the node name from the parsed tag content,

• it assumes there is only one string per tag
• thus mistakenly reads split character vector multiple times
• with the last character overwriting all the previous ones.
• then we ended up seeing partial (-) instead of complete node name (v- )

The quick fix will be hacking into graph:::graph_handler to concatenate these characters when they are coming from the same attribute of the same node.

— Reply to this email directly or view it on GitHubhttps://github.com/RGLab/flowWorkspace/issues/114#issuecomment-28945873 .

[mikejiang]

It is really the problem of XML parser. I think dot2gxl 's behaviour (encoding special character) on mac is more appropriate ( not sure why it doesn't do it for linux).