source: trunk/FACT++/src/csv2root.cc@ 19793

#include <random>

#include <boost/regex.hpp>
#include <boost/filesystem.hpp>
#include <boost/algorithm/string/join.hpp>

#include "tools.h"
#include "Time.h"
#include "Configuration.h"

#include <TROOT.h>
#include <TSystem.h>
#include <TFile.h>
#include <TTree.h>
#include <TError.h>
#include <TObjArray.h>

using namespace std;
namespace fs = boost::filesystem;

// ------------------------------------------------------------------------

void SetupConfiguration(Configuration &conf)
{
    po::options_description control("Root to SQL");
    control.add_options()
        ("file",           var<string>()->required(), "The csv input file")
        ("out,o",          var<string>(""),           "Output root file name")
        ("force,f",        po_switch(),               "Force overwrite if output file already exists.")
        ("update,u",       po_switch(),               "Update an existing file")
        ("tree,t",         var<string>("Events"),     "Name of the root tree to convert")
        ("compression,c",  var<uint16_t>(1),          "zlib compression level for the root file")
        ("no-header",      po_switch(),               "Use if the first line contains no header")
        ("dry-run",        po_switch(),               "Do not create or manipulate any output file")
        ;

    po::options_description split("Splitting options");
    split.add_options()
        ("split-sequence,S", vars<uint16_t>(),            "Split data sequentially into several trees/files (e.g. 1, 1, 2)")
        ("split-quantile,Q", vars<double>(),              "Split data randomly into several trees/files (e.g. 0.5, 1)")
        ("seed", var<uint64_t>(mt19937_64::default_seed), "Seed value in case of random split")
        ;

    po::options_description debug("Debug options");
    debug.add_options()
        ("verbose,v",      var<uint16_t>(1),          "Verbosity (0: quiet, 1: default, 2: more, 3, ...)")
        ;

    po::positional_options_description p;
    p.add("file", 1); // All positional options
    p.add("out",  1); // All positional options

    conf.AddOptions(control);
    conf.AddOptions(split);
    conf.AddOptions(debug);
    conf.SetArgumentPositions(p);
}

void PrintUsage()
{
    cout <<
        "csv2root - Reads data from a root tree and writes a csv file\n"
        "\n"
        "For convenience, this documentation uses the extended version of the options, "
        "refer to the output below to get the abbreviations.\n"
        "\n"
        "This is a general purpose tool to fill the contents of a root file into a database "
        "as long as this is technically possible and makes sense. Note that root can even "
        "write complex data like a TH1F into a database, this is not the purpose of this "
        "program.\n"
        "\n"
        "Each root tree has branches and leaves (the basic data types). These leaves can "
        "be read independently of the classes which were used to write the root file. "
        "The default tree to read from is 'Events' but the name can be overwritten "
        "using --tree. The default table name to fill the data into is identical to "
        "the tree name. It can be overwritten using --table.\n"
        "\n"
        "To get a list of the contents (keys and trees) of a root file, you can use --print-ls. "
        "The name of each column to which data is filled from a leave is obtained from "
        "the leaves' names. The leave names can be checked using --print-leaves. "
        "A --print-branches exists for convenience to print only the high-level branches. "
        "Sometimes these names might be quite unconvenient like MTime.fTime.fMilliSec or "
        "just MHillas.fWidth. To allow to simplify column names, regular expressions "
        "(using boost's regex) can be defined to change the names. Note that these regular "
        "expressions are applied one by one on each leaf's name. A valid expression could "
        "be:\n"
        "   --map=MHillas\\.f/\n"
        "which would remove all occurances of 'MHillas.f'. This option can be used more than "
        "once. They are applied in sequence. A single match does not stop the sequence.\n"
        "\n"
        "Sometimes it might also be convenient to skip a leaf. This can be done with "
        "the --ignore resource. If the given regular expresion yields a match, the "
        "leaf will be ignored. Note that the regular expression works on the raw-name "
        "of the leaf not the readily mapped SQL column names. Example:\n"
        "   --ignore=ThetaSq\\..*\n"
        "will skip all leaved which start with 'ThetaSq.'. This option can be used"
        "more than once.\n"
        "\n"
        "The data type of each column is kept as close as possible to the leaves' data "
        "types. If for some reason this is not wanted, the data type of the SQL column "
        "can be overwritten with --sql-type sql-column/sql-ytpe, for example:\n"
        "   --sql-type=FileId/UNSIGNED INT\n"
        "while the first argument of the name of the SQL column to which the data type "
        "should be applied. The second column is the basic SQL data type. The option can "
        "be given more than once.\n"
        "\n"
        "Database interaction:\n"
        "\n"
        "To drop an existing table, --drop can be used.\n"
        "\n"
        "To create a table according to theSQL  column names and data types, --create "
        "can be used. The query used can be printed with --print-create even --create "
        "has not been specified.\n"
        "\n"
        "To choose the columns which should become primary keys, use --primary, "
        "for example:\n"
        "   --primary=col1\n"
        "To define more than one column as primary key, the option can be given more than "
        "once. Note that the combination of these columns must be unique.\n"
        "\n"
        "All columns are created as NOT NULL as default. To force a database engine "
        "and/or a storage format, use --engine and --row-format.\n"
        "\n"
        "Usually, the INSERT query would fail if the PRIMARY key exists already. "
        "This can be avoided using the 'ON DUPLICATE KEY UPDATE' directive. With the "
        "--duplicate, you can specify what should be updated in case of a duplicate key. "
        "To keep the row untouched, you can just update the primary key "
        "with the identical primary key, e.g. --duplicate='MyPrimary=VALUES(MyPrimary)'. "
        "The --duplicate resource can be specified more than once to add more expressions "
        "to the assignment_list. For more details, see the MySQL manual.\n"
        "\n"
        "For debugging purpose, or to just create or drop a table, the final insert "
        "query can be skipped using --no-insert. Note that for performance reason, "
        "all data is collected in memory and a single INSERT query is issued at the "
        "end.\n"
        "\n"
        "Another possibility is to add the IGNORE keyword to the INSERT query by "
        "--ignore-errors, which essentially ignores all errors and turns them into "
        "warnings which are printed after the query succeeded.\n"
        "\n"
        "Using a higher verbosity level (-v), an overview of the written columns or all "
        "processed leaves is printed depending on the verbosity level. The output looks "
        "like the following\n"
        "   Leaf name [root data type] (SQL name)\n"
        "for example\n"
        "   MTime.fTime.fMilliSec [Long64_t] (MilliSec)\n"
        "which means that the leaf MTime.fTime.fMilliSec is detected to be a Long64_t "
        "which is filled into a column called MilliSec. Leaves with non basic data types "
        "are ignored automatically and are marked as (-n/a-). User ignored columns "
        "are marked as (-ignored-).\n"
        "\n"
        "A constant value for the given file can be inserted by using the --const directive. "
        "For example --const.mycolumn=42 would insert 42 into a column called mycolumn. "
        "The column is created as INT UNSIGNED as default which can be altered by "
        "--sql-type. A special case is a value of the form `/regex/format/`. Here, the given "
        "regular expression is applied to the filename and it is newly formated with "
        "the new format string. Uses the standard formatting rules to replace matches "
        "(those used by ECMAScript's replace method).\n"
        "\n"
        "Usually the previously defined constant values are helpful to create an index "
        "which relates unambiguously the inserted data to the file. It might be useful "
        "to delete all data which belongs to this particular file before new data is "
        "entered. This can be achieved with the `--delete` directive. It deletes all "
        "data from the table before inserting new data which fulfills the condition "
        "defined by the `--const` directives.\n"
        "\n"
        "The constant values can also be used for a conditional execution (--conditional). "
        "If any row with the given constant values are found, the execution is stopped "
        "(note that this happend after the table drop/create but before the delete/insert.\n"
        "\n"
        "To ensure efficient access for a conditonal execution, it makes sense to have "
        "an index created for those columns. This can be done during table creation "
        "with the --index option.\n"
        "\n"
        "To create the index as a UNIQUE INDEX, you can use the --unique option which "
        "implies --index.\n"
        "\n"
        "If a query failed, the query is printed to stderr together with the error message. "
        "For the main INSERT query, this is only true if the verbosity level is at least 2 "
        "or the query has less than 80*25 bytes.\n"
        "\n"
        "In case of success, 0 is returned, a value>0 otherwise.\n"
        "\n"
        "Usage: root2sql [options] -uri URI rootfile.root\n"
        "\n"
        ;
    cout << endl;
}

/*
void ErrorHandlerAll(Int_t level, Bool_t abort, const char *location, const char *msg)
{
    if (string(msg).substr(0,24)=="no dictionary for class ")
        return;
    if (string(msg).substr(0,15)=="unknown branch ")
        return;

    DefaultErrorHandler(level, abort, location, msg);
}*/


void AddTree(vector<TTree*> &ttree, TFile &file, const string &tree, bool update, int verbose)
{
    TTree *T = 0;
    if (update)
    {
        file.GetObject(tree.c_str(), T);
        if (T)
        {
            ttree.emplace_back(T);
            if (verbose>0)
                cout << "Updating tree: " << tree << endl;
        }
    }
    if (!T)
        ttree.emplace_back(new TTree(tree.c_str(), "csv2root"));
}

int main(int argc, const char* argv[])
{
    Time start;

    gROOT->SetBatch();
    //SetErrorHandler(ErrorHandlerAll);

    Configuration conf(argv[0]);
    conf.SetPrintUsage(PrintUsage);
    SetupConfiguration(conf);

    if (!conf.DoParse(argc, argv))
        return 127;

    // ----------------------------- Evaluate options --------------------------
    const string file            = conf.Get<string>("file");
    const string tree            = conf.Get<string>("tree");

    const bool force             = conf.Get<bool>("force");
    const bool update            = conf.Get<bool>("update");
    const bool dryrun            = conf.Get<bool>("dry-run");
    const bool noheader          = conf.Get<bool>("no-header");

    const uint16_t verbose       = conf.Get<uint16_t>("verbose");
//    const int64_t  first         = conf.Get<int64_t>("first");
//    const int64_t  max           = conf.Get<int64_t>("max");

    const uint16_t compression   = conf.Get<uint16_t>("compression");

    string out = conf.Get<string>("out");
    if (out.empty())
    {
        out = file;
        const auto p = out.find_last_of('.');
        if (p!=string::npos)
            out = string(out.substr(0, p))+".root";
    }

    // ----------------------------- Setup splitting ---------------------------

    vector<uint16_t> split_seq   = conf.Vec<uint16_t>("split-sequence");
    vector<double>   split_quant = conf.Vec<double>("split-quantile");

    if (!split_seq.empty() && !split_quant.empty())
        throw runtime_error("Only splitting by --split-sequence or --split-quantile is allowed.");

    const size_t num_split = split_seq.size()+split_quant.size()==0 ? 0 :
        ::max(split_seq.size(), split_quant.size()+1);

    map<size_t, size_t> split_lut;
    for (size_t i=0; i<split_seq.size(); i++)
    {
        const size_t sz = split_lut.size();
        for (size_t j=0; j<split_seq[i]; j++)
            split_lut.emplace(j+sz, i);
    }

    for (size_t i=0; i<split_quant.size(); i++)
        if (split_quant[i]<0 || split_quant[i]>=1)
            throw runtime_error("Splitting quantiles must be in the range [0;1)");

    for (size_t i=1; i<split_quant.size(); i++)
    {
        if (split_quant[i]<=split_quant[i-1])
            throw runtime_error("Splitting quantiles must be in increasing order.");
    }

    // -------------------------------------------------------------------------

    const uniform_real_distribution<double> distribution(0,1);
    mt19937_64 generator;
    generator.seed(conf.Get<uint64_t>("seed"));
    auto rndm = bind(distribution, generator);

    // -------------------------------------------------------------------------

    if (verbose>0)
    {
        cout << "\n-------------------------- Evaluating input ------------------------\n";
        cout << "Start Time: " << Time::sql << Time(Time::local) << endl;
    }


    // -------------------------------------------------------------------------

    cout << "Reading from '" << file << "'.\n";

    ifstream fin(file.c_str());
    if (!fin.good())
    {
        cerr << file << ": " << strerror(errno) << endl;
        return 1;
    }

    TString buf;
    buf.ReadLine(fin);
    if (!fin)
    {
        cerr << file << ": " << strerror(errno) << endl;
        return 2;
    }

    TObjArray *title = buf.Tokenize(" ");
    if (title->GetEntries()==0)
    {
        cerr << "First line empty." << endl;
        return 3;
    }

    if (title->At(0)->GetName()[0]=='#')
        title->RemoveAt(0);

    const auto numcol = title->GetEntries();

    if (verbose>0)
        cout << "Found " << numcol << " columns." << endl;

    if (noheader)
        fin.seekg(0);

    // -------------------------------------------------------------------------

    TString path(out.c_str());
    gSystem->ExpandPathName(path);

    if (!dryrun)
    {
        FileStat_t stat;
        const Int_t  exist  = !gSystem->GetPathInfo(path, stat);
        const Bool_t _write = !gSystem->AccessPathName(path,  kWritePermission) && R_ISREG(stat.fMode);

        if ((update && !exist) || (update && exist && !_write) || (force && exist && !_write))
        {
            cerr << "File '" << path << "' is not writable." << endl;
            return 4;
        }

        if (!update && !force && exist)
        {
            cerr << "File '" << path << "' already exists." << endl;
            return 5;
        }
    }

    TFile tfile(path, update?"UPDATE":(force?"RECREATE":"CREATE"), file.c_str(), compression);
    if (tfile.IsZombie())
    {
        cerr << "Opening file '" << path << "' failed." << endl;
        return 6;
    }

    if (verbose>0)
    {
        cout << "Opened root file '" << path << "'.\n";
        cout << "Writing to tree: " << tree << ".\n";
    }

    // -------------------- Configure branches of TTree ------------------------
    vector<TTree*> ttree;

    if (num_split==0)
        AddTree(ttree, tfile, tree, update, verbose);
    else
    {
        for (size_t i=0; i<num_split; i++)
            AddTree(ttree, tfile, tree+"["+to_string(i)+"]", update, verbose);
    }


    vector<float> vec(numcol);
    for (int i=0; i<numcol; i++)
    {
        string col = noheader ? Tools::Form("col%d", i) : title->At(i)->GetName();

        boost::regex rexpr(":");
        col = boost::regex_replace(col, rexpr, "");

        if (verbose>1)
            cout << "Column: " << col << '\n';

        for (auto it=ttree.begin(); it!=ttree.end(); it++)
            it[0]->Branch(col.c_str(), vec.data()+i);
    }

    delete title;

    // -------------------------------------------------------------------------

    size_t line = 0;

    while (1)
    {
        buf.ReadLine(fin);
        if (!fin)
            break;

        TObjArray *arr = buf.Tokenize(" ");
        if (arr->GetEntries()!=numcol)
        {
            cerr << "Column count mismatch in line " << line+1 << "!" << endl;
            return 6;
        }

        for (int i=0; i<numcol; i++)
        {
            try
            {
                vec[i] = stof(arr->At(i)->GetName());
            }
            catch (const exception &e)
            {
                cerr << "Conversion of '" << arr->At(i)->GetName() << "' failed!" << endl;
                return 7;
            }
        }

        delete arr;


        size_t index = 0;
        if (!split_lut.empty())
            index = split_lut[line % split_lut.size()];
        if (!split_quant.empty())
        {
            const float r = rndm();
            for (; r>=split_quant[index]; index++)
                if (index==split_quant.size())
                    break;
        }

        ttree[index]->Fill();
        line++;
    }

    if (verbose>0)
    {
        cout << line << " data rows read from file." << endl;
        for (size_t i=0; i<ttree.size(); i++)
            cout << ttree[i]->GetEntries() << " rows filled into tree #" << i << "." << endl;
    }

    for (auto it=ttree.begin(); it!=ttree.end(); it++)
        (*it)->Write("", TObject::kOverwrite);
    tfile.Close();

    if (verbose>0)
    {
        const auto sec = Time().UnixTime()-start.UnixTime();

        cout << Tools::Scientific(tfile.GetSize()) << "B written to disk.\n";
        cout << "File closed.\n";
        cout << "Execution time: " << sec << "s ";
        cout << "(" << Tools::Fractional(sec/line) << "s/row)\n";
        cout << "--------------------------------------------------------------" << endl;
    }

    return 0;
}