Skip to content

Instantly share code, notes, and snippets.

@carlosparadis

carlosparadis/socio_technical_analysis.r

Created April 2, 2017 07:42
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save carlosparadis/f7bb4b8875ae80b7fac8ed45088e1879 to your computer and use it in GitHub Desktop.
Save carlosparadis/f7bb4b8875ae80b7fac8ed45088e1879 to your computer and use it in GitHub Desktop.
Download ZIP
Sample R File
Raw
socio_technical_analysis.r
#! /usr/bin/env Rscript
## This file is part of Codeface. Codeface 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, version 2.
##
## This program 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 this program; if not, write to the Free Software
## Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
##
## Copyright 2016 by Mitchell Joblin <mitchell.joblin.ext@siemens.com>
## Copyright 2016 by Wolfgang Mauerer <wolfgang.mauerer@oth-regensburg.de>
## All Rights Reserved.
s <- suppressPackageStartupMessages
s(library(ggplot2))
s(library(igraph))
s(library(BiRewire))
s(library(GGally))
s(library(lubridate))
s(library(corrplot))
source("query.r")
source("utils.r")
source("config.r")
source("dependency_analysis.r")
source("process_dsm.r")
source("process_jira.r")
source("quality_analysis.r")
source("ml/ml_utils.r", chdir=T)
source("id_manager.r")
#################################################################################
plot.to.file <- function(g, outfile) {
g <- simplify(g,edge.attr.comb="first")
g <- delete.vertices(g, names(which(degree(g)<2)))
E(g)[is.na(E(g)$color)]$color <- "#0000001A"
pdf(file=outfile, width=10, height=10)
plot(g, layout=layout.kamada.kawai, vertex.size=2,
vertex.label.dist=0.5, edge.arrow.size=0.5,
vertex.label=NA)
dev.off()
}
motif.generator <- function(type, person.role, artifact.type, vertex.coding,
anti=FALSE) {
ensure.supported.artifact.type(artifact.type)
motif <- graph.empty(directed=FALSE)
if (type=="square") {
motif <- add.vertices(motif, 4)
motif <- add.edges(motif, c(1,2, 1,3, 2,4, 3,4))
if (anti) motif <- delete.edges(motif, c(1))
V(motif)$kind <- c(person.role, person.role, artifact.type,
artifact.type)
V(motif)$color <- vertex.coding[V(motif)$kind]
}
else if (type=="triangle") {
motif <- add.vertices(motif, 3)
motif <- add.edges(motif, c(1,2, 1,3, 2,3))
if (anti) motif <- delete.edges(motif, c(1))
V(motif)$kind <- c(person.role, person.role, artifact.type)
V(motif)$color <- vertex.coding[V(motif)$kind]
}
else {
motif <- NULL
}
return(motif)
}
preprocess.graph <- function(g, person.role) {
## Remove loops and multiple edges
g <- simplify(g, remove.multiple=TRUE, remove.loops=TRUE,
edge.attr.comb="first")
## Remove low degree artifacts
##artifact.degree <- degree(g, V(g)[V(g)$kind==artifact.type])
##low.degree.artifact <- artifact.degree[artifact.degree < 2]
##g <- delete.vertices(g, v=names(low.degree.artifact))
## Remove isolated developers
dev.degree <- degree(g, V(g)[V(g)$kind==person.role])
isolated.dev <- dev.degree[dev.degree==0]
g <- delete.vertices(g, v=names(isolated.dev))
return(g)
}
cor.mtest <- function(mat, conf.level = 0.95) {
mat <- as.matrix(mat)
n <- ncol(mat)
p.mat <- lowCI.mat <- uppCI.mat <- matrix(NA, n, n)
diag(p.mat) <- 0
diag(lowCI.mat) <- diag(uppCI.mat) <- 1
for (i in 1:(n - 1)) {
for (j in (i + 1):n) {
tmp <- cor.test(mat[, i], mat[, j], conf.level = conf.level,
method="spearman")
p.mat[i, j] <- p.mat[j, i] <- tmp$p.value
##lowCI.mat[i, j] <- lowCI.mat[j, i] <- tmp$conf.int[1]
##uppCI.mat[i, j] <- uppCI.mat[j, i] <- tmp$conf.int[2]
}
}
return(list(p.mat, lowCI.mat, uppCI.mat))
}
## Some helper functions to ensure that functions receive correct
## parameter values
ensure.supported.artifact.type <- function(artifact.type) {
if(!(artifact.type %in% c("function", "file", "feature"))) {
stop(str_c("Internal error: Artifact type ", artifact.type,
" is unsupported!"))
}
}
ensure.supported.dependency.type <- function(dependency.type) {
if(!(dependency.type %in% c("co-change", "dsm", "feature_call", "none"))) {
stop(str_c("Internal error: Dependency type ", dependency.type,
" is unsupported!"))
}
}
ensure.supported.quality.type <- function(quality.type) {
if(!(quality.type %in% c("corrective", "defect"))) {
stop(str_c("Internal error: Quality type ", quality.type,
" is unsupported!"))
}
}
ensure.supported.communication.type <- function(communication.type) {
if(!(communication.type %in% c("mail", "jira"))) {
stop(str_c("Internal error: Communication type ", communication.type,
" is unsupported!"))
}
}
## Compute communication relations between contributors
compute.communication.relations <- function(conf, communication.type,
jira.filename, start.date, end.date) {
ensure.supported.communication.type(communication.type)
if (communication.type=="mail") {
comm.dat <- query.mail.edgelist(conf$con, conf$pid, start.date, end.date)
colnames(comm.dat) <- c("V1", "V2", "weight")
} else if (communication.type=="jira") {
comm.dat <- load.jira.edgelist(conf, jira.filename, start.date, end.date)
}
if (is.null(comm.dat)) {
return(NULL)
}
comm.dat[, c(1,2)] <- sapply(comm.dat[, c(1,2)], as.character)
return(comm.dat)
}
## Compute entity-entity relations
compute.ee.relations <- function(conf, vcs.dat, start.date, end.date,
dependency.type, artifact.type,
dsm.filename, historical.limit) {
ensure.supported.dependency.type(dependency.type)
ensure.supported.artifact.type(artifact.type)
## Compute a list of relevant files (i.e., all files touched in the release
## range)
relevant.entity.list <- unique(vcs.dat$entity)
if (dependency.type == "co-change") {
start.date.hist <- as.Date(start.date) - historical.limit
end.date.hist <- start.date
commit.df.hist <- query.dependency(conf$con, conf$pid, artifact.type,
file.limit, start.date.hist,
end.date.hist)
commit.df.hist <- commit.df.hist[commit.df.hist$entity %in%
relevant.entity.list,]
## Compute co-change relationship
freq.item.sets <- compute.frequent.items(commit.df.hist)
## Compute an edgelist
dependency.dat <- compute.item.sets.edgelist(freq.item.sets)
names(dependency.dat) <- c("V1", "V2")
} else if (dependency.type == "dsm") {
dependency.dat <- load.sdsm(dsm.filename)
if (is.null(dependency.dat)) {
logwarning(str_c("Could not obtain any dependencies from the SDSM! ",
"Trying to continue without dependencies.\n",
"Is the implementation language supported?"), logger="conway")
return(data.frame())
}
dependency.dat <-
dependency.dat[dependency.dat[, 1] %in% relevant.entity.list &
dependency.dat[, 2] %in% relevant.entity.list,]
} else if (dependency.type == "feature_call") {
graph.dat <- read.graph(feature.call.filename, format="pajek")
V(graph.dat)$name <- V(graph.dat)$id
dependency.dat <- get.data.frame(graph.dat)
dependency.dat <-
dependency.dat[dependency.dat[, 1] %in% relevant.entity.list &
dependency.dat[, 2] %in% relevant.entity.list,]
names(dependency.dat) <- c("V1", "V2")
} else {
dependency.dat <- data.frame()
}
return(dependency.dat)
}
MOTIF.SEARCH.LIMIT=10*60 # Allow for trying up to 10 minutes of motif searching
do.motif.search <- function(motif, g, count.only=TRUE) {
## Compute the matching domains (i.e., which vertices in the pattern are
## allowed to be matched with which vertices in the larger graph)
dom <- lapply(V(motif)$color, function(col) { V(g)[V(g)$color==col] })
if (count.only) {
## Count subgraph isomorphisms in the collaboration graph with
## the given motif (this is faster than actually computing the subgraphs)
motif.count <- count_subgraph_isomorphisms(motif, g, method="lad", domain=dom,
induced=TRUE, time.limit=MOTIF.SEARCH.LIMIT)
return(motif.count)
}
motif.subgraphs <- subgraph_isomorphisms(motif, g, method="lad", domain=dom,
induced=TRUE, time.limit=MOTIF.SEARCH.LIMIT)
return(motif.subgraphs)
}
do.null.model <- function(g.bipartite, g.nodes, person.role, dependency.dat,
dependency.edgelist, comm.inter.dat, vertex.coding,
motif, motif.anti) {
## Rewire dev-artifact bipartite
g.bipartite.rewired <- birewire.rewire.bipartite(simplify(g.bipartite),
verbose=FALSE)
## Add rewired edges
gbr.df <- get.data.frame(g.bipartite.rewired)
g.null <- add.edges(g.nodes,
as.character(do.interleave(gbr.df$from, gbr.df$to)))
## Aritfact-artifact edges
if (nrow(dependency.dat) > 0) {
g.null <- add.edges(g.null, dependency.edgelist)
}
## Rewire dev-dev communication graph
g.comm <- graph.data.frame(comm.inter.dat)
g.comm.null <- birewire.rewire.undirected(simplify(g.comm),
verbose=FALSE)
## Ensure degree distribution is identical for graph and null model
if(!all(sort(as.vector(degree(g.comm.null))) ==
sort(as.vector(degree(g.comm))))) {
stop("Internal error: degree distribution not conserved!")
}
g.null <- add.edges(g.null,
as.character(with(get.data.frame(g.comm.null),
do.interleave(from, to))))
## Code and count motif
V(g.null)$color <- vertex.coding[V(g.null)$kind]
g.null <- preprocess.graph(g.null, person.role)
count.positive <- do.motif.search(motif, g.null)
count.negative <- do.motif.search(motif.anti, g.null)
ratio <- count.positive/(count.positive+count.negative)
res <- data.frame(count.type=c("positive", "negative", "ratio"),
count=c(count.positive, count.negative, ratio))
return(res)
}
## Generate a text string with some information about the range under analysis
gen.plot.info <- function(stats) {
return(str_c(stats$project, " (", stats$start.date, "--", stats$end.date,
")\n", "Devs: ", stats$num.devs,
" Funs: ", stats$num.functions,
" M: ", stats$num.motifs,
" A-M: ", stats$num.motifs.anti, sep=""))
}
do.quality.analysis <- function(conf, vcs.dat, quality.type, artifact.type, defect.filename,
start.date, end.date, communication.type, motif.type,
motif.subgraphs, motif.anti.subgraphs, df, stats, range.resdir) {
## Perform quality analysis
if (length(motif.subgraphs) == 0 || length(motif.anti.subgraphs) == 0) {
loginfo("Quality analysis lacks motifs or anti-motifs, exiting early")
return(NULL)
}
relevant.entity.list <- unique(vcs.dat$entity)
if (quality.type=="defect") {
quality.dat <- load.defect.data(defect.filename, relevant.entity.list,
start.date, end.date)
} else {
quality.dat <- get.corrective.count(conf$con, project.id, start.date,
end.date, artifact.type)
}
artifacts <- count(data.frame(entity=unlist(lapply(motif.subgraphs,
function(i) i[[3]]$name))))
anti.artifacts <- count(data.frame(entity=unlist(lapply(motif.anti.subgraphs,
function(i) i[[3]]$name))))
## Get file developer count
file.dev.count.df <- ddply(vcs.dat, .(entity),
function(df) data.frame(entity=unique(df$entity),
dev.count=length(unique(df$id))))
compare.motifs <- merge(artifacts, anti.artifacts, by='entity', all=TRUE)
compare.motifs[is.na(compare.motifs)] <- 0
names(compare.motifs) <- c("entity", "motif.count", "motif.anti.count")
artifacts.dat <- merge(quality.dat, compare.motifs, by="entity")
artifacts.dat <- merge(artifacts.dat, file.dev.count.df, by="entity")
## Add features
artifacts.dat$motif.percent.diff <- 2 * abs(artifacts.dat$motif.anti.count -
artifacts.dat$motif.count) /
(artifacts.dat$motif.anti.count + artifacts.dat$motif.count)
artifacts.dat$motif.ratio <- artifacts.dat$motif.anti.count /
(artifacts.dat$motif.count + artifacts.dat$motif.anti.count)
artifacts.dat$bug.density <- artifacts.dat$BugIssueCount /
(artifacts.dat$CountLineCode+1)
artifacts.dat$motif.count.norm <- artifacts.dat$motif.count /
artifacts.dat$dev.count
artifacts.dat$motif.anti.count.norm <- artifacts.dat$motif.anti.count /
artifacts.dat$dev.count
## Generate correlation plot (omitted correlation quantities: BugIsseChurn,
## IssueCommits, motif.count.norm, motif.anti.count.norm, motif.ratio, motif.percent.diff
corr.cols <- c("motif.ratio", "motif.percent.diff",
"dev.count", "bug.density",
"BugIssueCount", "Churn",
"CountLineCode")
corr.cols.labels <- c("M/A-M", "MDiff", "Devs", "BugDens",
"Bugs", "Churn", "LoC")
## Do not plot correlations for all quantities, but only for combinations
## that are of particular interest.
artifacts.subset <- artifacts.dat[, corr.cols]
colnames(artifacts.subset) <- corr.cols.labels
correlation.plot <- ggpairs(artifacts.dat,
columns=corr.cols,
lower=list(continuous=wrap("points",
alpha=0.33, size=0.5)),
upper=list(continuous=wrap('cor',
method='spearman')),
title=gen.plot.info(stats)) + theme_bw() +
theme(axis.title.x = element_text(angle = 90, vjust = 1,
color = "black"))
corr.mat <- cor(artifacts.subset, use="pairwise.complete.obs",
method="spearman")
corr.test <- cor.mtest(artifacts.subset)
## Write correlations and raw data to file
corr.plot.path <- file.path(range.resdir, "quality_analysis", motif.type,
communication.type)
dir.create(corr.plot.path, recursive=T, showWarnings=T)
pdf(file.path(corr.plot.path, "correlation_plot.pdf"), width=10, height=10)
print(correlation.plot)
dev.off()
pdf(file.path(corr.plot.path, "correlation_plot_color.pdf"),
width=7, height=7)
corrplot(corr.mat, p.mat=corr.test[[1]],
insig = "p-value", sig.level=0.05, method="pie", type="lower")
title(gen.plot.info(stats))
dev.off()
write.csv(artifacts.dat, file.path(corr.plot.path, "quality_data.csv"))
}
do.conway.analysis <- function(conf, global.resdir, range.resdir, start.date, end.date,
titandir) {
project.name <- conf$project
project.id <- conf$pid
dsm.filename <- file.path(titandir, "sdsm", "project.sdsm")
feature.call.filename <- "/home/mitchell/Documents/Feature_data_from_claus/feature-dependencies/cg_nw_f_1_18_0.net"
jira.filename <- file.path(global.resdir, "jira-comment-authors-with-email.csv")
defect.filename <- file.path(range.resdir, "time_based_metrics.csv")
## Analysis
motif.type <- list("triangle", "square")[[2]]
artifact.type <- list("function", "file", "feature")[[2]]
dependency.type <- list("co-change", "dsm", "feature_call", "none")[[1]]
quality.type <- list("corrective", "defect")[[2]]
communication.type <- list("mail", "jira")[[2]]
## Constants
person.role <- "developer"
file.limit <- 30
historical.limit <- ddays(365)
## Compute developer-artifact relationships that are obtained
## from the revision control system
vcs.dat <- query.dependency(conf$con, project.id, artifact.type, file.limit,
start.date, end.date, impl=FALSE, rmv.dups=FALSE)
vcs.dat$entity <- sapply(vcs.dat$entity,
function(filename) filename <- gsub("/", ".", filename, fixed=T))
vcs.dat$author <- as.character(vcs.dat$author)
## Determine which functions (node.function) and people (node.dev) contribute
## to the developer-artifact relationships
node.function <- unique(vcs.dat$entity)
node.dev <- unique(c(vcs.dat$author))
## Save to csv TODO: Why do we need to save that?
write.csv(vcs.dat, file.path(range.resdir, "commit_data.csv"))
## Compute various other relationships between contributors and/or entities
comm.dat <- compute.communication.relations(conf, communication.type,
jira.filename, start.date, end.date)
if (is.null(comm.dat)) {
loginfo(str_c("Conway analysis: No usable communication relationships available ",
"for time interval ", start.date, "--", end.date, ", exiting early", sep=""),
startlogger="conway")
return(NULL)
}
dependency.dat <- compute.ee.relations(conf, vcs.dat, start.date, end.date,
dependency.type, artifact.type,
dsm.filename, historical.limit)
## Generate a bipartite network that describes the socio-technical structure
## of a development project. This data structure is the core of the Conway
## analysis, and to make statements about relations between the technical
## and the social structure of a project.
## g.nodes is the basis for the graph it contains all relevant nodes,
## persons (developers) and artefacts. g.nodes has no edges.
g.nodes <- graph.empty(directed=FALSE)
g.nodes <- add.vertices(g.nodes, nv=length(node.dev),
attr=list(name=node.dev, kind=person.role,
type=TRUE))
g.nodes <- add.vertices(g.nodes, nv=length(node.function),
attr=list(name=node.function, kind=artifact.type,
type=FALSE))
## Graph g.bipartite based on g.nodes that contains developer-entity edges
vcs.edgelist <- with(vcs.dat, do.interleave(author, entity))
g.bipartite <- add.edges(g.nodes, vcs.edgelist, attr=list(color="#00FF001A"))
## Create a graph g that is based on g.bipartite and that additionally
## captures developer-developer communication
## * First, remove persons that don't appear in VCS data, and transfer
## to remaining edges of g.bipartite into g
g <- graph.empty(directed=FALSE)
comm.inter.dat <- comm.dat[comm.dat$V1 %in% node.dev &
comm.dat$V2 %in% node.dev,]
if (dim(comm.inter.dat)[1] == 0) {
loginfo("No overlap between VCS and communication data, exiting analysis early!",
logger="conway")
return(NULL)
}
comm.edgelist <- as.character(with(comm.inter.dat,
do.interleave(V1, V2)))
g <- add.edges(g.bipartite, comm.edgelist, attr=list(color="#FF00001A"))
## * Second, add entity-entity edges
if(nrow(dependency.dat) > 0) {
dependency.edgelist <- as.character(with(dependency.dat,
do.interleave(V1, V2)))
g <- add.edges(g, dependency.edgelist)
}
## * Third, remove some undesired outlier contributions of the graph
## that complicate further processing.
g <- preprocess.graph(g, person.role)
## * Fourth, define a numeric encoding scheme for vertices, and color
## the different vertices
vertex.coding <- c()
vertex.coding[person.role] <- 1
vertex.coding[artifact.type] <- 2
V(g)$color <- vertex.coding[V(g)$kind]
## * Last, save the resulting graph for external processing
## (TODO: This should go into the database)
write.graph(g, file.path(range.resdir, "network_data.graphml"), format="graphml")
################# Analyse the socio-technical graph ##############
## Generate motif and anti-motif that we want to find in the data
motif <- motif.generator(motif.type, person.role, artifact.type,
vertex.coding)
motif.anti <- motif.generator(motif.type, person.role, artifact.type,
vertex.coding, anti=TRUE)
## Find motif and anti-motifs in the collaboration graph
logdevinfo("Searching for motif and anti-motif", logger="conway")
motif.subgraphs <- do.motif.search(motif, g, count.only=FALSE)
motif.count <- length(motif.subgraphs)
motif.anti.subgraphs <- do.motif.search(motif.anti, g, count.only=FALSE)
motif.anti.count <- length(motif.anti.subgraphs)
## Compute a null model
niter <- 100
logdevinfo("Computing null models", logger="conway")
motif.count.null <- mclapply(seq(niter), function(i) {
do.null.model(g.bipartite, g.nodes, person.role, dependency.dat,
dependency.edgelist, comm.inter.dat, vertex.coding,
motif, motif.anti)
}, mc.cores=2)
null.model.dat <- do.call(rbind, motif.count.null)
null.model.dat[null.model.dat$count.type=="positive", "empirical.count"] <-
motif.count
null.model.dat[null.model.dat$count.type=="negative", "empirical.count"] <-
motif.anti.count
null.model.dat[null.model.dat$count.type=="ratio", "empirical.count"] <-
motif.count/(motif.count+motif.anti.count)
logdevinfo("Cleaning up null models", logger="conway")
## When we did neither find motifs nor anti-motifs, set the fraction
## M/(M+A-M) to one (instead of NaN), since there are exactly as many motifs as
## anti-motifs. NaNs would also break the plotting code below.
if (any(is.nan(null.model.dat$count))) {
null.model.dat[is.nan(null.model.dat$count),]$count <- 1
}
if (any(is.nan(null.model.dat$empirical.count))) {
null.model.dat[is.nan(null.model.dat$empirical.count),]$empirical.count <- 1
}
logdevinfo("Computing null model stats", logger="conway")
stats <- list(num.devs=length(node.dev), num.functions=length(node.function),
num.motifs=motif.count, num.motifs.anti=motif.anti.count,
start.date=start.date, end.date=end.date, project=conf$project)
## Visualise the results in some graphs
networks.dir <- file.path(range.resdir, "motif_analysis", motif.type,
communication.type)
dir.create(networks.dir, recursive=TRUE, showWarnings=TRUE)
logdevinfo("Writing null model data", logger="conway")
write.table(null.model.dat, file=file.path(networks.dir, "raw_motif_results.txt"))
## Visualise the null model
labels <- c(negative = "Anti-Motif", positive = "Motif", ratio = "Ratio")
p.null <- ggplot(data=null.model.dat, aes(x=count)) +
geom_histogram(aes(y=..density..), colour="black", fill="white", bins=30) +
geom_point(aes(x=empirical.count), y=0, color="red", size=5) +
geom_density(alpha=.2, fill="#AAD4FF") +
facet_wrap(~count.type, scales="free", labeller=labeller(count.type=labels)) +
xlab("Count") + ylab("Density [a.u.]") + ggtitle(gen.plot.info(stats))
ggsave(plot=p.null,
filename=file.path(networks.dir, "motif_null_model.pdf"),
width=8, height=4)
## Compute the communication degree distribution
p.comm <- ggplot(data=data.frame(degree=degree(graph.data.frame(comm.dat))),
aes(x=degree)) +
geom_histogram(aes(y=..density..), colour="black", fill="white", bins=30) +
geom_density(alpha=.2, fill="#AAD4FF") + xlab("Node degree") + ylab("Density [a.u.]") +
ggtitle(gen.plot.info(stats))
ggsave(plot=p.comm,
filename=file.path(networks.dir, "communication_degree_dist.pdf"),
width=7, height=8)
## Plot the complete network
plot.to.file(g, file.path(networks.dir, "socio_technical_network.pdf"))
do.quality.analysis(conf, vcs.dat, quality.type, artifact.type, defect.filename,
start.date, end.date, communication.type, motif.type,
motif.subgraphs, motif.anti.subgraphs, df, stats, range.resdir)
}
######################### Dispatcher ###################################
config.script.run({
conf <- config.from.args(positional.args=list("project_resdir", "range_resdir",
"start_date", "end_date"),
require.project=TRUE)
global.resdir <- conf$project_resdir
range.resdir <- conf$range_resdir
start.date <- conf$start_date
end.date <- conf$end_date
titandir <- file.path(range.resdir, "titan")
logdevinfo(paste("Directory for storing conway results is", range.resdir),
logger="conway")
dir.create(global.resdir, showWarnings=FALSE, recursive=TRUE)
dir.create(range.resdir, showWarnings=FALSE, recursive=TRUE)
if (conf$profile) {
## R cannot store line number profiling information before version 3.
if (R.Version()$major >= 3) {
Rprof(filename="ts.rprof", line.profiling=TRUE)
} else {
Rprof(filename="ts.rprof")
}
}
do.conway.analysis(conf, global.resdir, range.resdir,
start.date, end.date, titandir)
})
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment