TEXPROFILE(1) Displaying TeX profiles TEXPROFILE(1)

texprofile - Displaying the profile data collected by texprof.

texprofile [Options] inputfile

texprofile reads the binary inputfile as produced by texprof. The extension .tprof of the input file can be omitted. It analyses the data and presents the results in a variety of tables, either optimized for human readability or as CSV (comma separated values) files for further processing.

The output is written to the standard output stream where it can be redirected to a file or viewed immediately.

There are four types of options: general options, table options, selection options, and formatting options.

The general options:

-? -h --help

Display a short help text and exit the program.

Display the version information and exit.

The table options determine the tables that will be displayed. These options all use upper case letters. If no table option is given, only some global information is shown.


Show the table of the top 10 input lines.

Show the table of the macro call graph.

Show the table of times per TeX command.

Show the table of times per input line.

Show the table of raw time stamps.

Show the table of macro stack changes.

Show the table of all input files.

Show the table of all macros called.

Show important tables (equal to -TGFC) tables.

Do not show the table of global information.

The selection options allow to reduce the amount of data that is shown in the table by omitting table entries that contribute little to the overall runtime.


Do not show information for table entries with cumulative time below n percent. The default is 1.0 percent.

Limit the number of input lines shown in the table of the ``Top Ten'' to n. This option is only useful with the -T option.

The formatting option determine the formatting of the information in the tables as well as the selection of information that is shown in the tables.


Add the macros file and line numbers after the macro name. This option is useful to distinguish two macros that share the same name.

Optimize output for machine readability. When optimized for human readability, times are rounded and given in a short form adding a unit identifier: s for seconds, ms for milliseconds, us for micro seconds, and ns for nanoseconds. This format is inconvenient if you want to import the data for example into a spread-sheet for further processing. With the -m option times are simply given in nano seconds without units. Similar the combined time for the total time used for a macro and the time used as a sub-macro from the call graph is given in a single column with a slash separating both numbers is nice for human readers. With the -m option both numbers will simply be given in separate columns.

Show the time stamp numbers. This option is useful together with the -R option, if you import the table into a database or spread-sheet. If you sort the data by various properties into different categories, the time stamp number can be used to keeping the entries within a category sorted by time.

Show the changes of the macro stack. This option is useful with the -R option if you want to see the changes in the macro stack (see -S) together with the timing information in a single table.

While texprof is running, every TeX input file is given a unique file number. The file name alone is often not unique because two files in different directories might have the same file name. Displaying the full file name with the complete path is however often not very convenient. So if you observe the same file name together with different file numbers, you can use the option -F to get a table of all file numbers and their full file names.

But even within the same directory, TeX can read different files with the same name during one run: TeX can open a file for writing, write content to the file, close it, open it for reading, read it and close it again. Then TeX might repeat this process a second time, or multiple times, reusing the same file name over and over again. texprof will assign a new file number to this file each time it is opened for reading. So you can tell from the sequence of file numbers which one was the first, the second, or the third.

Last not least, there are some special file numbers:

0 - unknown
If the file is unknown, which should rarely happen, the file number 0 is used.
1- system
texprof will map time intervals that are used to execute certain system routines to the ``system'' pseudo file using the line numbers to identify the specific routine like producing the output DVI file (shipout), breaking a paragraph into lines (linebrk), or breaking the document into pages (buildpg). These times do not depend on the use of macros but simply on the size of the document.
2 - terminal
TeX commands entered on the command line or interactively are mapped to file number 2.

If the option -m is given times are given as nanoseconds. Otherwise, times are rounded to at most 4 digits precission and displayed with a unit: seconds (s), milliseconds (ms), microseconds (um), or nanoseconds (ns).

Macro names are shown with the leading backslash. Since macro names are often not unique, the command line option -i can be used to show after the macro name in square brackets the file number and the line number where the macro is defined. Unless two macros with the same name are defined in the same file and line this is sufficient to uniquely identify a macro. A macro defined with ``let'' results in a complet copy of the original macro. Therefore it will not reference the file and line of the ``let'' command but the same file and line as the original macro.

Let's assume that you issue the command texprof -prof hello.tex. This will run texprof on the input file hello.tex with option -prof. The program texprof will load the plain TeX format and then process hello.tex to produce hello.log and hello.dvi. It will execute exactly the same steps that TeX would execute if you had issued the command tex hello.tex.

In addition to hello.log and hello.dvi, texprof will also produce the file hello.tprof containing time measurements made while texprof was running. The option -prof will switch on the gathering of timing data as soon as texprof enters its main control procedure; The file hello.tprof will contain a time measurement, called a time stamp, for every command that TeX executed while processing the input. A binary format is used to store all that data in a simple and compact form. Still the file hello.tprof might become very big.

The program texprofile is used to extract and analyse the data contained in hello.tprof and produce useful output. Here are some examples:


Without further options texprofile will write some general information to the standard output, like the total time measured, the number of samples, the average time per sample, etc. The general information is always given unless explicitely disabled with the -N option or with the -m option.

With the option -T, texprofile will map each time stamp to a specific line of input, add up the time intervalls for each input line separately, and output a table showing the ten lines that have the highest cummulative time. The table has the following seven columns:
1. file
The first column contains the input file number as explained above. The input file name if shown in column 7.
2. line
The line number.
3. percent
The time spent in this line and file as a percentage of the total time measured as given as part of the general information.
4. absolute
The absolute time spent in this line and file.
5. count
The number of times the execution path entered this line. Note, that a macro call usually redirects the execution path to another line from where the execution will return after the macro call has completed. Reentering the line after such an excursion to an other line will cause this counter to be incremented. In summary, this counter might reflect the number of partial executions of a line not the number of full executions of the entire line.
6. average
The average time spent in the line is simply computed by dividing the value in column 4 by the value in column 5.
7. file
The input file name. The corresponding file number is shown in column 1.

With the option -G, texprofile will map each time stamp either to file input or to a macro body. The table shown is divided into several sections, the first section is devoted to file input, each of the following sections is devoted to a specific macro.
Each section starts with a header. The header of the first section is ``File'' the header of a later section is the macro name.
The first line after the header gives the total time spent in the section in two different formats:
1. time
Column 1 gives it as an absolut time.
3. percentage
Column 3 gives it as a percentage of the total time measured.

For the first section with the header ``File'', the absolute time will be equal to the total time measured because texprof did spent all the time processing the file hello. And consequently the value in column 3 will be 100%.
To accomplish a task, a macro usually calls other macros that we call child macros in the following. The following lines in the table will give a breakdown of the time shown in the first line. The breakdown starts with a line showing the time spent in the section excluding the time spent in child macros. It shows in column 4 the number of times the macro was called.

1. time
Column 1 gives the absolut time spent in the section excluding the time spent in child macros
3. percentage
Column 3 gives the time from column 1 as a percentage of the total time spent in this section as given in the previous line.
4. count/total
Column 4 gives the number of times the section was called.

The lines that follow in the table show the time spent in one of the child macros. Column 2, 4, and 5 need some explanation.

2. loop
The only case where column 2 is not empty is the case of a recursive macro. A recursive macro is a macro that along the chain of macro calls eventually calls itself creating a recursive loop. At this point, a macro becomesits own descendant and at the same time its own ancestor.
Therefore texprofile maintains for each child macro two accumulators for the elapsed time: For the time shown in column 2 labeled ``loop'', texprofile adds up the time differences observed at the return of a child macro. For the time shown in the column labeled ``time'' and ``percent'', it subtracts from the time differences observed at the return of a child macro all those time differences that were already added to one of the other lines in the time breakdown: the macro itself or one of the other child macros. So the times shown in column 1 of line 2 and the following lines will add up to the time shown in column 1 of the line 1; and the percentages shown in column 3 of line 2 and the following lines will add up to 100%. The time shown in column 3 will show show the total time needed to accomplish the sub task assigned to the respective child macro.
4. count/total
Column 4 shows two counts n/m for the macro named in column 5. m is the total number of calls to the macro and n is the number of calls as a child macro in the current section. The number n will always be less or equal to m.
5. child
Column 5 shows the name of the child macro as explained above.

This table will contain the same data as the previous example but this time the -m option will optimize the output for machine readability.
*
There are no column headers.
*
The times in column 1 ``time'' and column 2 ``loop'' are given in nanoseconds without a unit identifier.
*
The numbers n/m in column 4 are now shown in two separate columns 4 and 5; the macro name moves from column 5 to column 6.

With the option -F, texprofile outputs the table of all of TeX's input files using 5 columns:
1. file
The file number as explained above.
2. lines
The difference between the highest and the lowest line number found in the profile data for this file. This is usually only a subset of all the lines of the file.
3. percent
The percentage of the total time measured that is attibuted to the file.
4. time
The absolute time that is attibuted to the file.
5. name
The full file name of the file.

With the option -C, texprofile outputs the table of all TeX commands executed while profiling using 6 columns:
1. cmd
The command code used internaly by TeX. Usually the same number is used by TeX for several closely related tasks. There is a special command code 101 which is used to for the time that is spend on system routines that are mapped to the system file as explained above.
2. time
The total time used for the command.
3. percent
The percentage of the total time measured used for the command.
4. count
The number of times this command was executed.
5. average
The average time needed to execute the command. This is simpy the value in column 2 divided by the value in column 4.
6. name
A verbal description of the command or commands that share this command code.

With the options -R -m, texprofile outputs the table of raw time measurements as observed by texprof optimized for machine readability. The option -n adds a column for the number of each time measurements. Because of the -m option, the table has no column headers. The table has 7 columns:
1. number
The number of the time measurment. These numbers are strictly increasing but not necessarily consecutive. They can be used to keep the measurements sorted in the order in which they were made.
2. file
The file number as explained above.
3. line
The line number as explained above.
4. time
The time interval in nano seconds.
5. command
The command name as given in column 6 of the previous example.
6. level
The nesting level of the macro call stack.
7. macro
The macro name.

This table contains all the timing information gathered during the run of texprof. Some information about macro calls is contained in column 6. More information about macro calls could be added by using the -s option. The information about macro returns can be obtained from column 6. The table can be imported to a spread-sheet or a database for further analysis.

If the last command in a macro body is a macro call, we call this a tail call. If such a tail call reads ahead to scan the following input for possible arguments, the look-ahead mechaism of TeX might push further macros or new input files on TeX's input stack. These entries will remain on top of TeXprof's macro nesting stack, even if TeX backs up all these tokens on its input stack. This can cause an attribution of runtime to those entries as sub entries of the tail call. If this explanation sounds complicated to you, it is because the situation is indeed complicated.

texprof(1)

Version: 1.1 of 28-1-2025

(C)2024, Martin Ruckert,
Hochschule München, Lothstrasse 64, D-80335 München
texprofile is distributed with TeX Live. More recent versions might be found on GitHub https://github.com/ruckertm/HINT.
This program can be redistributed and/or modified under the terms of the MIT/X11 license.

Martin Ruckert
Munich University of Applied Sciences
Email: martin.ruckert@hm.edu

28-1-2025 Displaying TeX profiles