library(ggplot2)
library(dplyr)
library(readr)
library(stringr)
library(tidyr)

# Usage: Rscript combined_fault_probability.r exp_abspath1 ... resultsdata_file
#
# Divides by the faultspace area instead of by a marker total, which makes the
# running modes comparable: raw counts scale with how long WAMR runs, so they
# say more about execution length than about susceptibility.
#
# Each segment is P[outcome] for a uniformly random single-bit flip in the
# traced fault space, so a bar's height is P[anything goes wrong].
#
# Replaces combined_fault_composition.r, which divided by the marker total and
# was therefore this chart with every bar rescaled to 100%.

args <- commandArgs(trailingOnly = TRUE)
if (length(args) < 2) {
  stop("Need at least 2 experiments")
}

csv_suffix <- if (grepl("\\.csv$", args[length(args)])) {
  args[length(args)]
} else {
  "resultsdata.csv"
}
exp_args <- if (grepl("\\.csv$", args[length(args)])) {
  args[-length(args)]
} else {
  args
}

extract_info <- function(path) {
  dir_name <- basename(path)
  match <- str_match(
    dir_name,
    "^\\d{2}-\\d{2}_\\d{2}-\\d{2}-\\d{2}_(.+?)_(c|aot|interp)_"
  )
  if (is.na(match[1, 1])) {
    warning(paste("Could not parse:", dir_name))
    return(NULL)
  }
  list(base_name = match[1, 2], variant = match[1, 3], path = path)
}

# Load data
all_data <- data.frame()

for (arg in exp_args) {
  info <- extract_info(arg)
  if (is.null(info)) {
    next
  }

  csv_file <- file.path(info$path, csv_suffix)
  if (!file.exists(csv_file)) {
    warning(paste("Missing:", csv_file))
    next
  }
  df <- read_csv(csv_file, col_types = cols())
  df$base_name <- info$base_name
  df$variant <- info$variant
  all_data <- bind_rows(all_data, df)
}

# TODO: Finally put all the bullshit before this in some shared space

if (nrow(all_data) == 0) {
  stop("No data loaded")
}

marker_order <- c(
  "OK_MARKER",
  "DETECTED_MARKER",
  "GROUP1_MARKER",
  "TRAP",
  "TIMEOUT",
  "WRITE_TEXTSEGMENT",
  "ACCESS_OUTERSPACE",
  "FAIL_MARKER"
)

# Don't merge GROUP1_MARKER into TRAP for this chart
# Also keep the OK_MARKERs, so the "sum to 100%" is accurate
probability <- all_data |>
  group_by(base_name, variant, benchmark) |>
  mutate(frac = faults / sum(faults, na.rm = TRUE)) |>
  ungroup()

# Don't print alphabetically
probability$resulttype <- factor(probability$resulttype, levels = marker_order)
probability$variant <- factor(
  probability$variant,
  levels = c("c", "aot", "interp")
)

plot <- ggplot(
  probability,
  aes(x = variant, y = frac, fill = resulttype)
) +
  geom_col() +
  facet_grid(benchmark ~ base_name) +
  scale_y_continuous(labels = scales::percent) +
  labs(
    title = "Fault Probability per Fault Space",
    x = NULL,
    y = "Probability of Failure",
    fill = "Fault Type"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(size = 13, face = "bold"),
    axis.text.x = element_text(angle = 45, hjust = 1)
  )

out_suffix <- gsub("^resultsdata|\\.csv$", "", csv_suffix)
filename <- paste0("injections/fault_probability", out_suffix, ".svg")
ggsave(filename, plot = plot, width = 13, height = 8)