mw-lifecycle-analysis/commit_analysis/plotting/relevance-plot.R

#library(tidyverse)
library(dplyr)
library(lubridate)
library(tidyr)
library(purrr)
library(stringr)

https_commit_fp <- "/mmfs1/gscratch/comdata/users/mjilg/mw-repo-lifecycles/case2/mediawiki_core_commits.csv"

contains_http_but_not_url <- function(text) {
  if (is.na(text)) {
    return(FALSE)
  }
  # Split text by whitespace and check each word
  words <- str_split(text, "\\s+")[[1]]
  for (word in words) {
    if (!str_detect(word,"://")){
      #http
      if (str_detect(word, "http")){
        return(TRUE)
      }
      if (str_detect(word, "login")){
        return(TRUE)
      }
      if (str_detect(word, "ssl")){
        return(TRUE)
      }
      if (str_detect(word, "tls")){
        return(TRUE)
      }
      if (startsWith(word, "cert") && !startsWith(word, "certain")){
        return(TRUE)
      }
    }
  }
  return(FALSE)
}

transform_relevant_commit_data <- function(filepath){
  #basic, loading in the file 
  df = read.csv(filepath, header = TRUE) 
  temp_df <- df
  dir_path = dirname(filepath)
  file_name = basename(filepath)
  
  
  # TODO: this is project/event specific
  event_date <- as.Date("2013-08-28")
  #event_date <- as.Date("2013-07-01")
  #event_date <- as.Date("2013-04-25")
  #event_date <- as.Date("2012-12-11")
  
  # isolate project id
  project_id <- sub("_.*$", "", file_name)
  
  #make sure the dates are formatted correctly and state the project_id
  df <- df |>
    mutate(commit_date = ymd_hms(commit_date)) |>
    mutate(project_id = project_id)
  
  #get information about project age either in the "present" 
  #or at the time of first commit
  oldest_commit_date <- min(as.Date(df$commit_date))
  project_age <- as.numeric(as.Date("2025-02-10") - oldest_commit_date)
  
  #add that to the data 
  df <- df |> 
    mutate(age = project_age)
  
  #drop out data from ''before'' the release process 
  df <- df |>
    filter(commit_date >= as.Date("2011-09-03"))
  
  #we are looking at weekly data, 6m before and 6m after
  #start_date <- event_date %m-% months(6)
  #calculated_start_date <- event_date %m-% months(12)
  start_date <- as.Date('2011-09-03')
  end_date <-as.Date('2015-10-03')
  
  #getting the relative weeks to the publication date
  relative_week <- function(date, ref_date) {
    as.integer(as.numeric(difftime(date, ref_date, units = "days")) %/% 7)
  }
  
  df <- df |>
    mutate(relative_week = relative_week(commit_date, event_date))|>
    mutate(mentions_http = if_else(sapply(message, contains_http_but_not_url), TRUE, FALSE)) 
  #  arrange(relative_week) |>
  #  group_by(author_email) |>
  #  mutate(new_author = ifelse(row_number() <= 5, 1, 0),
  #         new_author_wmf = if_else(grepl("@wikimedia", author_email), new_author, 0),
  #         new_author_unaff = if_else(!grepl("@wikimedia", author_email), new_author, 0)) |>
  #  ungroup()
  
  # cut the df to all before 06-01-2015
  # group by author_email 
  # list all author_emails with >5 commits 
  # for big df: if author not in the list, 'new' author
  old_author_list <- df |>
    filter(commit_date > as.Date("2011-09-03") & commit_date < as.Date("2013-08-01"))|>
    group_by(author_email) |>
    summarise(commit_count = n()) |>
    filter(commit_count > 5) |>
    pull(author_email)
  
  # Label authors as 'new' if they are not in the old_author_list
  df <- df |>
    mutate(new_author = ifelse(author_email %in% old_author_list, 0, 1), 
           new_author_wmf = if_else(grepl("@wikimedia", author_email), 
                                    new_author, 0),
           new_author_unaff = if_else(!grepl("@wikimedia", author_email) &
                                        !grepl("l10n-bot@translatewiki.net|tools.libraryupgrader@tools.wmflabs.org", author_email),
                                      new_author, 0))
  
  
  #filler for when there are weeks without commits
  all_weeks <- seq(relative_week(start_date, event_date), relative_week(end_date, event_date))
  complete_weeks_df <- expand.grid(relative_week = all_weeks,
                                   project_id = project_id,
                                   age = project_age)
  
  
  #for each week, get the list of unique authors that committed
  #cumulative_authors <- df %>%
  #  arrange(relative_week) %>%
  #  group_by(relative_week) %>%
  #  summarize(cumulative_author_emails = list(unique(author_email)), .groups = 'drop')
  #same for each committer
  #cumulative_committers <- df %>%
  #  arrange(relative_week) %>%
  #  group_by(relative_week) %>%
  #  summarize(cumulative_committer_emails = list(unique(committer_email)), .groups = 'drop')
  
  #now cut out the commit data that we don't care about 
  df <- df |>
    filter(author_email != "jenkins-bot@gerrit.wikimedia.org") |>
    filter(author_email != "l10n-bot@translatewiki.net") |>
    filter(author_email != "tools.libraryupgrader@tools.wmflabs.org") 
  
  #in order:
  # - we group by project, week, ages
  # - and we summarize commit and authorship details
  # - we then fill in information for missingness
  # - and add in vars for before/after
  # - and weekly index
  weekly_commits <- df |>
    group_by(project_id, relative_week, age) |>
    summarise(commit_count = n(),
              author_emails = list(unique(author_email)),
              committer_emails = list(unique(committer_email)),
              mediawiki_dev_commit_count = sum(grepl("@users.mediawiki.org", author_email)),
              wikimedia_commit_count = sum(grepl("@wikimedia", author_email)),
              wikia_commit_count = sum(grepl("@wikia-inc.com", author_email)),
              bot_commit_count = sum(grepl("l10n-bot@translatewiki.net|tools.libraryupgrader@tools.wmflabs.org", author_email)),
              wmf_new_commit_count = sum(new_author_wmf),
              unaff_new_commit_count = sum(new_author_unaff),
              relevant_commits = sum(mentions_http),
              .groups = 'drop') |>
    right_join(complete_weeks_df, by=c("relative_week", "project_id", "age")) |>
    replace_na(list(commit_count = 0)) |>
    replace_na(list(wikimedia_commit_count = 0)) |>
    replace_na(list(l10n_commit_count = 0)) |>
    replace_na(list(jenkins_commit_count = 0)) |>
    replace_na(list(mediawiki_dev_commit_count = 0)) |>
    replace_na(list(wikia_commit_count = 0)) |>
    replace_na(list(bot_commit_count = 0)) |>
    replace_na(list(wmf_new_commit_count = 0)) |>
    replace_na(list(unaff_new_commit_count = 0)) |>
    replace_na(list(relevant_commits = 0)) |>
    mutate(before_after = if_else(relative_week < 0, 0, 1)) |>
    select(-author_emails, -committer_emails)
  
  weekly_commits <- weekly_commits |>
    filter(relative_week >= (-103) & relative_week <= 109 )
  
  #gracefully exit
  return(weekly_commits)
}

transformed <- transform_relevant_commit_data(https_commit_fp)

transformed$irrelevant_commit = transformed$commit_count - transformed$relevant_commits

long_df <- transformed|>
  tidyr::pivot_longer(cols = c(irrelevant_commit, relevant_commits),
                      names_to = "commit_relevance",
                      values_to = "lengthened_commit_count")

relevant_https_commits <- long_df |>
  ggplot(aes(x=relative_week,
             y=lengthened_commit_count, 
            fill=commit_relevance)) +
  geom_col(position = "dodge", width = 0.7) +
  labs(x = "Relative Week", y = "Commits", fill="Commit relevance") +
  scale_fill_manual(values = c("irrelevant_commit" = "#E1BE6A",  # Color for "Returning Contributors"
                               "relevant_commits" = "#40B0A6"),   
                    labels = c("irrelevant_commit" = "Irrelevant Commits",
                               "relevant_commits" = "Relevant Commits")
  ) +
  ggtitle("Commits to MW-Core, 09-03-2011 to 10-03-2015, by relevance to HTTP/s feature deployments") +
  theme_bw() +
  theme(legend.position = "top")

relevant_https_commits

ggsave(filename = "ww-c2c3-relevance-viz.png", plot = relevant_https_commits, width = 12, height = 9, dpi = 800)