Debian Codegraph @

docs/manual/andi-manual.tex

raw
1 \documentclass[a4paper,
2 10pt,
3 english,
4 DIV=12,
5 BCOR=8mm]{scrbook}
6 \usepackage[utf8x]{inputenc}
7 \usepackage{babel}
8 \usepackage{listings}
9 \usepackage{xcolor}
10 \usepackage{hyperref}
11 \usepackage{siunitx}
12 \usepackage[T1]{fontenc}
13 \usepackage{isodate}
14 \usepackage{graphicx}
15 \usepackage{amsthm}
16 \usepackage{acronym}
17 \usepackage{amssymb}
18 \usepackage{caption}
19 \usepackage{subcaption}
20 \usepackage{xspace}
21 \usepackage{microtype}
22
23 \bibliographystyle{alpha}
24
25 \DeclareSIUnit\byte{B}
26 \DeclareSIUnit\basepairs{bp}
27 \DeclareSIUnit\bit{bit}
28
29 \definecolor{oceangreen}{cmyk}{1,.0,.20,.78}
30 \addtokomafont{sectioning}{\rmfamily\color{oceangreen}}
31
32 \definecolor{bluekeywords}{rgb}{0.13,0.13,1}
33 \definecolor{greencomments}{rgb}{0,0.5,0}
34 \definecolor{turqusnumbers}{rgb}{0.17,0.57,0.69}
35 \definecolor{redstrings}{rgb}{0.5,0,0}
36 \definecolor{lightgray}{rgb}{0.9,0.9,0.9}
37
38 \usepackage{libertine}
39 \fontfamily{libertine}
40 \selectfont
41 %\usepackage[scaled]{berasans}
42
43 \newcommand{\thymine}{\textsc{m}\oldstylenums{2}\xspace}
44 \newcommand{\local}{\textsc{m}\oldstylenums{1}\xspace}
45 \newcommand{\algo}[1]{\textsc{{#1}}}
46 \newcommand{\andi}{\algo{andi}\xspace}
47 \newcommand{\word}[1]{\textsf{\small#1}}
48 \newcommand{\wchar}[1]{\textsf{\small#1}}
49 \newcommand{\eco}{\textsc{eco}\oldstylenums{29}\xspace}
50 \newcommand{\pneu}{\textsc{Pneu}\oldstylenums{3085}\xspace}
51
52 \include{version}
53
54 % Todos at the margin
55 \newcommand{\todo}[1]{
56 \marginpar{\fbox{\begin{minipage}{0.9\marginparwidth}
57 \scriptsize\sloppy\raggedright #1
58 \end{minipage}}}
59 }
60
61
62 \newtheorem{definition}{Definition}
63
64
65 \lstset{backgroundcolor=\color{lightgray}}
66
67 \lstdefinestyle{shell}{
68 language=bash,
69 columns=flexible,
70 xleftmargin=12pt,
71 xrightmargin=12pt,
72 breaklines=true,
73 basicstyle=\small\ttfamily,
74 morekeywords={make, tar, git, sudo, andi, time, man, head, cut, fneighbor,
75 fretree, figtree, brew, aura, autoreconf, ls},
76 % literate={~} {$\sim$}{1}
77 }
78
79 \lstset{style=shell}
80
81 \title{Documentation of \algo{andi}}
82 \subtitle{Rapid Estimation of Evolutionary Distances between Genomes\\ {\small\url{https://github.com/EvolBioInf/andi}}}
83 \author{Fabian Klötzl\\ \href{mailto:[email protected]}{[email protected]}}
84 \date{Version \version, \isodate\today \\
85 \vspace*{2cm}
86 \centering\includegraphics[width=0.8\textwidth]{andi_labels.pdf}}
87
88 \begin{document}
89
90 \maketitle
91
92 \section*{Abstract}
93 This is the documentation of the \andi program for estimating the evolutionary distance between closely related genomes. These distances can be used to rapidly infer phylogenies for big sets of genomes. Because \andi does not compute full alignments, it is so efficient that it scales well up to thousands of bacterial genomes.
94
95 This is scientific software. Please cite our paper \cite{andi} if you use \andi in your publication. Also refer to the paper for the internals of \andi. Additionally, there is a Master's thesis with even more in depth analysis of \andi \cite{kloetzl}.
96
97 \vspace*{1cm}
98 \section*{License}
99 This document is release under the Creative Commons Attribution Share-Alike license. This means, you are free to copy and redistribute this document. You may even remix, tweak and build upon this document, as long as you credit me for the work I've done and release your document under the identical terms. The full legal code is available online: {\small\url{https://creativecommons.org/licenses/by-sa/4.0/legalcode}}.
100
101 \tableofcontents
102
103 \chapter{Installation} %%%%%
104
105 \section{Package Manager}
106
107 The easiest way to install \andi is via a package manager. This also handles all dependencies for you.
108
109
110 \noindent Debian and Ubuntu:
111
112 \begin{lstlisting}
113 ~ % sudo apt-get install andi
114 \end{lstlisting}
115
116 \noindent macOS with homebrew:
117
118 \begin{lstlisting}
119 ~ % brew tap brewsci/bio
120 ~ % brew install andi
121 \end{lstlisting}
122
123 \noindent ArchLinux AUR package with aura:
124
125 \begin{lstlisting}
126 ~ % aura -A andi
127 \end{lstlisting}
128
129 \andi is intended to be run in a \algo{Unix} commandline such as \lstinline$bash$ or \lstinline$zsh$. All examples in this document are also intended for that environment. You can verify that \andi was installed correctly by executing \lstinline$andi -h$. This should give you a list of all available options (see Section~\ref{sec:options}).
130
131 \section{Source Package} \label{sub:regular}
132
133 To build \andi from source, download the latest \href{https://github.com/EvolBioInf/andi/releases}{release} from GitHub. Please note, that \andi requires the \algo{Gnu Scientific Library} and \algo{libdivsufsort}\footnote{\url{https://github.com/y-256/libdivsufsort}} for optimal performance \cite{divsufsort}.
134
135 Once you have downloaded the package, unzip it and change into the newly created directory.
136
137 \begin{lstlisting}
138 ~ % tar -xzvf andi-0.14.tar.gz
139 ~ % cd andi-0.14
140 \end{lstlisting}
141
142 \noindent Now build and install \andi.
143
144 \begin{lstlisting}
145 ~/andi-0.14 % ./configure
146 ~/andi-0.14 % make
147 ~/andi-0.14 % sudo make install
148 \end{lstlisting}
149
150 \noindent This installs \andi for all users on your system. If you do not have root privileges, you will find a working copy of \andi in the \lstinline$src$ subdirectory. For the rest of this documentation, it is assumed, that \andi is in your \textdollar\lstinline!PATH!.
151
152 Now \andi should be ready for use. Try invoking the help.
153
154 \begin{lstlisting}
155 ~/andi-0.14 % ~/andi
156 Usage: andi [OPTIONS...] FILES...
157 FILES... can be any sequence of FASTA files.
158 Use '-' as file name to read from stdin.
159 Options:
160 -b, --bootstrap=INT Print additional bootstrap matrices
161 --file-of-filenames=FILE Read additional filenames from FILE; one per line
162 -j, --join Treat all sequences from one file as a single genome
163 -l, --low-memory Use less memory at the cost of speed
164 -m, --model=MODEL Pick an evolutionary model of 'Raw', 'JC', 'Kimura', 'LogDet'; default: JC
165 -p FLOAT Significance of an anchor; default: 0.025
166 --progress=WHEN Print a progress bar 'always', 'never', or 'auto'; default: auto
167 -t, --threads=INT Set the number of threads; by default, all processors are used
168 --truncate-names Truncate names to ten characters
169 -v, --verbose Prints additional information
170 -h, --help Display this help and exit
171 --version Output version information and acknowledgments
172 \end{lstlisting}
173
174 \noindent \andi also comes with a man page, which can be accessed via \lstinline$man andi$. % But once you are done with this documentation, you will require it scarcely.
175
176 \section{Installing from Git Repository}
177
178 To build \andi from the \algo{Git} repo, you will also need the \algo{autotools}. Refer to your OS documentation for installation instructions. Once done, execute the following steps.
179
180 \begin{lstlisting}
181 ~ % git clone [email protected]:EvolBioInf/andi.git
182 ~ % cd andi
183 ~/andi % autoreconf -fi -Im4
184 \end{lstlisting}
185
186 \noindent Continue with the \algo{Gnu} trinity as described in Section~\ref{sub:regular}.
187
188
189 \chapter{Usage} %%%%%
190
191 The input sequences for \andi should be in \algo{Fasta} format. Any number of files can be passed. Each file may contain more than one sequence.
192
193 \begin{lstlisting}
194 ~ % andi S1.fasta S2.fasta
195 2
196 S1 0.0000 0.0979
197 S2 0.0979 0.0000
198 \end{lstlisting}
199
200 If no file argument is given, \andi reads the input from \algo{stdin}. This makes it convenient to use in \algo{Unix} pipelines.
201
202 \begin{lstlisting}
203 ~ % cat S1.fasta S2.fasta | andi
204 2
205 S1 0.0000 0.0979
206 S2 0.0979 0.0000
207 \end{lstlisting}
208
209 The output of \andi is a matrix in \algo{Phylip} style: On the first line the number of compared sequences is given, \lstinline!2! in our example. Then the matrix is printed, where each line is preceded by the name of the $i$th sequence. Note that the matrix is symmetric and the main diagonal contains only zeros. The numbers themselves are evolutionary distances, estimated from substitution rates.
210
211
212 \section{Input} \label{sec:join}
213
214 As mentioned before, \andi reads in \algo{Fasta} files. It recognizes only the four standard bases and is case insensitive (RegEx: \lstinline![acgtACGT]!). All other residue symbols are excluded from the analysis and \andi prints a warning, when this happens.
215
216 If instead of distinct sequences, a \algo{Fasta} file contains contigs belonging to a single taxon, \andi will treat them as a unit when switched into \algo{join} mode. This can be achieved by using the \lstinline!-j! or \lstinline!--join! command line switch.
217
218 \begin{lstlisting}
219 ~ % andi --join E_coli.fasta Shigella.fasta
220 [Output]
221 \end{lstlisting}
222
223 When the \algo{join} mode is active, the file names are used to label the individual sequences. Thus, in \algo{join} mode, each genome has to be in its own file, and furthermore, at least one filename has to be given via the command line.
224
225 If not enough file names are provided, \andi will try to read sequences from the standard input stream. This behaviour can be explicitly triggered by passing a single dash (\lstinline$-$) as a file name, which is useful in pipelines.
226
227 If \andi seems to take unusually long, or requires huge amounts of memory, then you might have forgotten the \algo{join} switch. This makes \andi compare each contig instead of each genome, resulting in many more comparisons! Since version 0.12 \andi produces a progressmeter on the standard error stream. \andi tries to be smart about when to show or hide the progress bar. You can manually change this behaviour using the \lstinline!--progress! option.
228
229 Starting with version 0.11 \andi supports an extra way of input. Instead of passing file names directly to \andi via the commandline arguments, the file names may also be read from a file itself. Using this new \lstinline$--file-of-filenames$ argument can work around limitations imposed be the shell.
230
231 The following three snippets have the same functionality.
232
233 \begin{lstlisting}
234 ~ % andi --join *.fasta
235 [Output]
236 \end{lstlisting}
237
238 \begin{lstlisting}
239 ~ % ls *.fasta > filenames.txt
240 ~ % andi --join --file-of-filenames filenames.txt
241 [Output]
242 \end{lstlisting}
243
244 \begin{lstlisting}
245 ~ % ls *.fasta | andi --join --file-of-filenames -
246 [Output]
247 \end{lstlisting}
248
249 \section{Output}
250
251 The output of \andi is written to \lstinline$stdout$. This makes it easy to use on the command line and within shell scripts. As seen before, the matrix, computed by \algo{andi}, is given in \algo{Phylip} format \cite{phylip}.
252
253 \begin{lstlisting}
254 ~ % cat S1.fasta S2.fasta | andi
255 2
256 S1 0.0000 0.0979
257 S2 0.0979 0.0000
258 \end{lstlisting}
259
260 If the computation completed successfully, \andi exits with the status code 0. Otherwise, the value of \lstinline$errno$ is used as the exit code. \andi can also produce warnings and error messages for the user's convenience. These messages are printed to \lstinline$stderr$ and thus do not interfere with the normal output.
261
262 \section{Options} \label{sec:options}
263
264 \andi takes a small number of commandline options, of which even fewer are of interest on a day-to-day basis. If \lstinline$andi -h$ displays a \lstinline$-t$ option, then \andi was compiled with multi-threading support (implemented using \algo{OpenMP}). By default, \andi uses all available processors. However, to restrict the number of threads, use \lstinline$-t$.
265
266 \begin{lstlisting}
267 ~ % time andi ../test/1M.1.fasta -t 1
268 2
269 S1 0.0000 0.0995
270 S2 0.0995 0.0000
271 ./andi ../test/1M.1.fasta 0,60s user 0,01s system 99% cpu 0,613 total
272 ~ % time andi ../test/1M.1.fasta -t 2
273 2
274 S1 0.0000 0.0995
275 S2 0.0995 0.0000
276 ./andi ../test/1M.1.fasta -t 2 0,67s user 0,03s system 195% cpu 0,362 total
277 \end{lstlisting}
278
279 In the above examples the runtime dropped from \SI{0.613}{\second}, to \SI{0.362}{\second} using two threads. Giving \andi more threads than input genomes leads to no further speed improvement. \, The other important option is \lstinline$--join$ (see Section~\ref{sec:join}).
280
281 By default, the distances computed by \andi are \emph{Jukes-Cantor} corrected \cite{jukescantor}. Other evolutionary models are also implemented (Kimura \cite{kimura}, LogDet \cite{logdet}, raw). The \lstinline$--model$ parameter can be used to switch between them.
282
283 Since version 0.9.4 \andi includes a bootstrapping method. It can be activated via the \lstinline$--bootstrap$ or \lstinline$-b$ switch. This option takes a numeric argument representing the number of matrices to create. The output can then be piped into \algo{phylip}. For more information on computing support values from distance matrices see \cite{afra}.
284
285 \begin{lstlisting}
286 ~ % andi -b 2 ../test/1M.1.fasta
287 2
288 S1 0.0000 0.1067
289 S2 0.1067 0.0000
290 2
291 S1 0.0000 0.1071
292 S2 0.1071 0.0000
293 \end{lstlisting}
294
295 The original \algo{phylip} only supports distance matrices with names no longer than ten characters. However, this sometimes leads to problems with long accession numbers. Starting with version 0.11 \andi prints the full name of a sequence, even if it is longer than ten characters. If your downstream tools have trouble with this, use \lstinline$--truncate-names$ to reimpose the limit.
296
297 Also new in version 0.11 is the \lstinline$--file-of-filenames$ option. See Section~\ref{sec:join} for details.
298
299 \section{Example: \algo{eco29}}
300
301 Here follows a real-world example of how to use \algo{andi}. It makes heavy use of the commandline and tools like \algo{Phylip}. If you prefer \algo{R}, check out this excellent blog post by Kathryn Holt.\footnote{\url{http://holtlab.net/2015/05/08/r-code-to-infer-tree-from-andi-output/}}
302
303 As a data set we use \algo{eco29}; 29 genomes of \textit{E. Coli} and \textit{Shigella}. You can download the data from here: {\small{\url{http://guanine.evolbio.mpg.de/andi/eco29.fasta.gz}}}. The genomes have an average length of 4.9~million nucleotides amounting to a total \SI{138}{\mega\byte}.
304
305 \algo{eco29} comes a single \algo{fasta} file, where each sequence is a genome. To calculate their pairwise distances, enter
306
307 \begin{lstlisting}
308 ~ % andi eco29.fasta > eco29.mat
309 andi: The input sequences contained characters other than acgtACGT. These were automatically stripped to ensure correct results.
310 \end{lstlisting}
311
312 \noindent The \algo{eco29} data set includes non-canonical nucleotides, such as \word{Y}, \word{N}, and \word{P}, which get stripped from the input sequences. The resulting matrix is stored in \lstinline$eco29.mat$; Here is a small excerpt:
313
314 \begin{lstlisting}
315 ~ % head -n 5 eco29.mat | cut -d ' ' -f 1-5
316 29
317 gi|563845 0.0000e+00 1.8388e-02 1.8439e-02 2.6398e-02
318 gi|342360 1.8388e-02 0.0000e+00 4.4029e-04 2.6166e-02
319 gi|300439 1.8439e-02 4.4029e-04 0.0000e+00 2.6123e-02
320 gi|261117 2.6398e-02 2.6166e-02 2.6123e-02 0.0000e+00
321 \end{lstlisting}
322
323 \noindent From this we compute a tree via neighbor-joining using a \algo{Phylip} wrapper called \algo{Embassy}.\footnote{\url{http://emboss.sourceforge.net/embassy/\#PHYLIP}}
324
325 \begin{lstlisting}
326 ~ % fneighbor -datafile eco29.mat -outfile eco29.phylipdump
327 \end{lstlisting}
328 \noindent To make this tree easier to read, we can midpoint-root it.
329 \begin{lstlisting}
330 ~ % fretree -spp 29 -intreefile eco29.treefile -outtreefile eco29.tree <<EOF
331 M
332 X
333 Y
334 R
335 EOF
336 \end{lstlisting}
337
338 \noindent The file \lstinline$eco29.tree$ now contains the tree in Newick format. This can be plotted using \cite{figtree}
339
340 \begin{lstlisting}
341 ~ % figtree eco29.tree &
342 \end{lstlisting}
343
344 \noindent to yield
345
346 \begin{figure}[h]
347 \centering\includegraphics[width=0.8\textwidth]{andi_labels.pdf}
348 \end{figure}
349
350
351 \chapter{Warnings and Errors}
352
353 Here be an explanation of all possible errors. Other errors may occur and are due to the failure of underlying functions (e.\,g.~\lstinline$read(3)$). All warning messages are printed to \lstinline$stderr$. Most errors are non-recoverable and will result in \andi exiting with a non-zero state.
354
355 \section{Sequence Related Messages}
356
357 \subsection*{Unexpected Character}
358
359 \andi is pretty pedantic about the formatting of \algo{FASTA} files. If you violate the syntax, \andi will print the file name, the line and the problematic character. These errors are non-recovering, meaning no further sequences are read from the invalid file. The checks are implemented by the \href{https://github.com/kloetzl/pfasta}{\algo{pfasta}} library.
360
361
362 \subsection*{Non acgtACGT Nucleotides Stripped}
363
364 Our models of genome evolution (JC, Kimura) only work on the four canonical types of nucleotides. All others are stripped from the sequences. This can be ignored in most cases.
365
366 \subsection*{Too Short Sequence}
367
368 \andi was designed for big data sets of whole genomes. On short sequences the distance estimates are inaccurate. Use a multiple sequence alignment instead.
369
370 \subsection*{Too Long Sequence}
371
372 \algo{libdivsufsort} limits the length of a sequence to 31 bits. That count includes the reverse complement. So the technical limit for a sequence analysis is $2^{30} = 1.073.741.824$. Unfortunately, that excludes (full) human and mice genomes. Per-chromosome analysis works just fine.
373
374 \subsection*{Empty Sequence}
375
376 One of the given sequences contained either no nucleotides at all, or only non-canonical ones.
377
378 \subsection*{Less than two sequences given}
379
380 As \andi tries to compare sequences, at least two need to be supplied. Note that \andi may have regarded some of your given sequences as unusable.
381
382 \subsection*{Maximum Number of Sequences}
383
384 The maximum number of sequences \andi can possible compare is huge (roughly $457.845.052$). I doubt anyone will ever reach that limit. Please send me a mail, if you do.
385
386 \section{Technical Messages}
387
388 \subsection*{Out of Memory}
389
390 If \andi runs out of memory, it gives up. Either free memory, run \andi on a bigger machine, try the \lstinline$--low-memory$ mode or reduce the number of threads.
391
392 \subsection*{RNG allocation}
393
394 Some technical thing failed. If it keeps failing repeatedly, file a bug.
395
396 \subsection*{Bootstrapping failed}
397
398 This should not happen.
399
400 \subsection*{Failed index creation}
401
402 This should not happen, either.
403
404 \subsection*{Skipped and ignored Arguments}
405
406 Some command line parameters of \andi require arguments. If these are not of the expected type, a warning is given. See Section~\ref{sec:options} for their correct usage.
407
408
409 \section{Output-related Warnings}
410
411 As the input sequences get more evolutionary divergent, \andi finds less homologous anchors. With less anchors, less nucleotides are considered homologous between two sequences. If no anchors are found, comparison fails and \lstinline!nan! is printed instead. See our paper and especially Figure~2 for details.
412
413 \subsection*{NaN}
414
415 No homologous sections were found. Your sequences are very divergent ($d>0.5$) or sprout a lot of indels that make comparison difficult.
416
417 \subsection*{Little Homology}
418
419 Very few anchors were found and thus only a tiny part of the sequences is considered homologous. Expect that the given distance is erroneous.
420
421 \subsection*{Too long name}
422
423 If you added the \lstinline$--truncate-names$ switch and an input name is longer than ten characters, you will receive this warning.
424
425 \chapter{DevOps} %%%%%
426
427 \andi is written in C/C++; mostly C99 with some parts in C++11. The sources are released on \algo{GitHub} as \emph{free software} under the \textsc{Gnu General Public License version~3} \cite{GPL}. Prebundled packages using \algo{autoconf} are also available, with the latest release being {\version} at the time of writing.
428
429 If you are interested in the internals of \algo{andi}, consult the paper \cite{andi} or my Master's thesis \cite{kloetzl}. Both explain the used approach in detail. The latter emphasizes the used algorithms, data structures and their efficient implementation.
430
431 \section{Dependencies}
432
433 Here is a complete list of dependencies required for developing \algo{andi}.
434
435 \begin{itemize}
436 \item A C and a C++11 compiler,
437 \item the \algo{autotools},
438 \item the \algo{Gnu Scientific Library},
439 \item \algo{Pdflatex} with various packages for the manual,
440 \item \algo{Git},
441 \item \algo{glib2} for the unit tests,
442 \item \algo{doxygen},
443 \item and \algo{libdivsufsort}.
444 \end{itemize}
445
446
447 \section{Code Documentation}
448
449 \emph{Every} function in \andi is documented using \algo{doxygen} style comments. To create the documentation run \lstinline$make code-docs$ in the main directory. You will then find the documentation under \lstinline$./docs$.
450
451
452 \section{Unit Tests}
453
454 The unit tests are located in the \andi repository under the \lstinline$./test$ directory. Because they require \algo{glib2}, and a C++11 compiler, they are deactivated by default. To enable them, execute
455
456 \begin{lstlisting}
457 ~/andi % ./configure --enable-unit-tests
458 \end{lstlisting}
459
460 \noindent during the installation process. You can then verify the build via
461
462 \begin{lstlisting}
463 ~/andi % make check
464 \end{lstlisting}
465
466 \noindent The unit tests are also checked each time a commit is sent to the repository. This is done via \algo{TravisCI}.\footnote{\url{https://travis-ci.org/EvolBioInf/andi}} Thus, a warning is produced, when the builds fail, or the unit tests did not run successfully. Currently, the unit tests cover more than 75\% of the code. This is computed via the \algo{Travis} builds and a service called \algo{Coveralls}.\footnote{\url{https://coveralls.io/r/EvolBioInf/andi}}
467
468 \section{Known Issues}
469
470 These minor issues are known. I intend to fix them, when I have time.
471
472 \begin{enumerate}
473 \item This code will not work under Windows. At two places Unix-only code is used: filepath-separators are assumed to be \lstinline$/$ and file-descriptors are used for I/O.
474 \item Unit tests for the bootstrapped matrices are missing.
475 \item Cached intervals are sometimes not “as deep as they could be”. If that got fixed \lstinline$get_match_cache$ could bail out on \lstinline$ij.lcp < CACHE_LENGTH$. However the \lstinline$esa_init_cache$ code is the most fragile part and should be handled with care.
476 \end{enumerate}
477
478
479 \section{Creating a Release}
480
481 A release should be a stable version of \andi with significant improvements over the last version. dotdot releases should be avoided.
482
483 %\subsection{Preparing a new Release}
484
485 Once \andi is matured, the new features implemented, and all tests were run, a new release can be created. First, increase the version number in \lstinline$configure.ac$. Commit that change in git, and tag this commit with \lstinline$vX.y$. Tags should be annotated and signed, if possible. This manual then needs manual rebuilding.
486
487 Ensure that \andi is ready for packaging with \algo{autoconf}.
488
489 \begin{lstlisting}
490 ~ % make distcheck
491 make dist-gzip am__post_remove_distdir='@:'
492 make[1]: Entering directory `/home/kloetzl/Projects/andi'
493 if test -d "andi-0.9.1-beta"; then find "andi-0.9.1-beta" -type d ! -perm -200 -exec chmod u+w {} ';' && rm -rf "andi-0.9.1-beta" || { sleep 5 && rm -rf "andi-0.9.1-beta"; }; else :; fi
494 test -d "andi-0.9.1-beta" || mkdir "andi-0.9.1-beta"
495 (cd src && make top_distdir=../andi-0.9.1-beta distdir=../andi-0.9.1-beta/src \
496 am__remove_distdir=: am__skip_length_check=: am__skip_mode_fix=: distdir)
497
498 ... Loads of output ...
499
500 =================================================
501 andi-0.9.1-beta archives ready for distribution:
502 andi-0.9.1-beta.tar.gz
503 =================================================
504 \end{lstlisting}
505
506 If the command does not build successfully, no tarballs will be created. This may necessitate further study of \algo{autoconf} and \algo{automake}.
507
508 Also verify that the recent changes did not create a performance regression. This includes testing both ends of the scale: \eco and \pneu. Both should be reasonable close to previous releases.
509
510 Create another commit, where you set the version number to the next release (e.\,g., \lstinline$vX.z-beta$). This assures that there is only one commit and build with that specific version.
511
512 \backmatter
513 %\addcontentsline{toc}{chapter}{Bibliography}
514 \bibliography{references}
515
516 \end{document}