[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[GNUnet-SVN] [taler-schemafuzz] branch master updated: rewrote so paragr
From: |
gnunet |
Subject: |
[GNUnet-SVN] [taler-schemafuzz] branch master updated: rewrote so paragrphs that were unclear |
Date: |
Fri, 31 Aug 2018 15:59:27 +0200 |
This is an automated email from the git hooks/post-receive script.
erwan-ulrich pushed a commit to branch master
in repository schemafuzz.
The following commit(s) were added to refs/heads/master by this push:
new 074ed00 rewrote so paragrphs that were unclear
074ed00 is described below
commit 074ed00c931d6f352c466fa2258755861e293c69
Author: Feideus <address@hidden>
AuthorDate: Fri Aug 31 15:59:21 2018 +0200
rewrote so paragrphs that were unclear
---
docs/Documentation.pdf | Bin 953766 -> 958332 bytes
docs/Documentation.tex | 149 +++++++++++++++++++++++++++----------------
docs/PersonnalExperience.tex | 10 ++-
3 files changed, 102 insertions(+), 57 deletions(-)
diff --git a/docs/Documentation.pdf b/docs/Documentation.pdf
index f645a17..d8af16d 100644
Binary files a/docs/Documentation.pdf and b/docs/Documentation.pdf differ
diff --git a/docs/Documentation.tex b/docs/Documentation.tex
index 5746f20..1b6bd7e 100644
--- a/docs/Documentation.tex
+++ b/docs/Documentation.tex
@@ -50,7 +50,7 @@ Traditional fuzzing is defined as "testing an automated
software testing techniq
-This quote is very well illustrated by the following example :
+This quote is well illustrated by the following example :
\begin{quotation}
Lets consider an integer in a program, which stores the result of a user's
choice between 3 questions. When the user picks one, the choice will be 0, 1 or
2. Which makes three practical cases. But what if we transmit 3, or 255 ? We
can, because integers are stored a static size variable. If the default switch
case hasn't been implemented securely, the program may crash and lead to
"classical" security issues: (un)exploitable buffer overflows, DoS, ...
\end{quotation}
@@ -64,7 +64,7 @@ However, SchemaFuzz is a database oriented fuzzer. This means
that it focuses on
This tool is meant to help developers, maintainers and more generically anyone
that makes use of data coming from a database under his influence in their
task. A good way to sum up the effect of this tool is to compare it with an
"cyber attack simulator".
This means that the idea behind it is to emulate the damage that an attacker
may cause subtly or not to a database he illegally gained privileges on. This
might in theory go from a simple boolean flip (subtle modifications) to
removing/adding content to purely and simply destroying or erasing all the
content of the database.
-SchemaFuzz focuses on the first part : modification of the content of the
database by single small modification that may or may not overlap. These
modifications may be very aggressive of very subtle.
+SchemaFuzz focuses on the first part : modification of the content of the
database by single small modification that may or may not overlap. These
modifications may be aggressive of subtle.
It is interesting to point out that this last point also qualifies SchemaFuzz
as a good "database structural flaw detector".
That is to say that errors typically triggered by a poor management of a
database (wrong data type usage, incoherence between database structure and use
of the content etc ...) might also appear clearly during the execution.
\subsection{Perimeter}
@@ -76,9 +76,9 @@ The resulting software will generate a group of human
readable reports on each m
\caption{Shows the nature of the code for every distinct
component. The slice size is a rough estimation.}
\end{figure}
\subsection{When to use it}
-SchemaFuzz is a very useful tool for anyone trying secure a piece of software
that uses database resources. The target software should be GDB(introduce GDB)
compatible and the DBMS(introduce acronym) has to grant access to the target
database through credentials passed as argument to this tool.
+SchemaFuzz is a useful tool for anyone trying secure a piece of software
that uses database resources. The target software should be GDB(introduce GDB)
compatible and the DBMS(introduce acronym) has to grant access to the target
database through credentials passed as argument to this tool.
----It is very strongly advice to use a copy of the target database rather than
on the production material. Doing so may result in the database being corrupted
and not usable for any useful mean.
+---It is strongly advice to use a copy of the target database rather than on
the production material. Doing so may result in the database being corrupted
and not usable for any useful mean.
\clearpage
@@ -94,7 +94,7 @@ The majority of this project is built on top of this already
existing code and i
This organization will be detailed and discussed in the following sections.
\subsection{SchemaSpy legacy/meta data extraction}
-SchemaSpy source code has provided the meta data extraction routine. The only
job of this routine is to initialize the connection to the database and
retrieve its meta data at the very beginning of the execution (before any
actual SchemaFuzz code is run). These meta data include data types, table and
table column names, views and foreign/primary key constraints. Having this pool
of meta data under the shape of Java objects allows the main program to
properly frame what the possibilities [...]
+SchemaSpy source code has provided the meta data extraction routine. The only
job of this routine is to initialize the connection to the database and
retrieve its meta data at the beginning of the execution (before any actual
SchemaFuzz code is run). These meta data include data types, table and table
column names, views and foreign/primary key constraints. Having this pool of
meta data under the shape of Java objects allows the main program to properly
frame what the possibilities are [...]
\clearpage
@@ -145,30 +145,15 @@ It also holds the informations concerning the result of
the injection in the sha
\caption{Structure of a Mutation}
\end{figure}
-\bigskip
-
- \paragraph{Choosing pattern}
-For each iteration of the main loop, a modification has to be picked up as the
next step in the fuzzing process. This is done by considering the current state
of the tree.
-Three parallel code paths can be triggered from this point.
- \begin{itemize}
- \item{Continue on the current branch of the
tree (triggered if the last mutation scored better than its parent)}
- \item{Pick an existing branch in the tree and
grow it (triggered if the last mutation scored worse than its parent on a 50/50
chance with the next bullet)}
- \item{Start a new branch (triggered if the last
mutation scored worse than its parent on a 50/50 chance with the previous
bullet)}
-
-\begin{figure}[h!]
-\centering
-\includegraphics[width=\textwidth]{pickingPaternDiagram.pdf}
-\caption{picking Pattern schema}
-\end{figure}
+\bigskip
- \end{itemize}
A branch is a succession of mutation that share the same database row as their
modification target.
-The heuristics determining the next mutation's modification are still very
primitive and will be thinly justed in futures versions.
+The heuristics determining the next mutation's modification are still
primitive and will be thinly justed in futures versions.
\paragraph{Creating malformed data}
%%ne veux rien dire.
As the goal of running this tool is to submit unexpected or invalid data to
the target software it is necessary to understand what t
Fuzzing a complex type such a timestamps variables has nothing to do with
fuzzing a trivial boolean. In practice, a significant part o
-and this matter could absolutely be the subject of a more abstract work. We
focused here on a very simple approach (as a first step).
+and this matter could absolutely be the subject of a more abstract work. We
focused here on a simple approach (as a first step).
After retrieving the current row being fuzzed (may it be a new row or a
previously fuzzed row), the algorithm explores the different
The algorithm then builds the possible modification for each of the fields for
the current row.
At the moment, the supported types are : % add a list of the supported types.
@@ -196,12 +181,12 @@ The possible modifications that this tool can produce at
the moment are : \\ % a
\item Set date to $00/00/0000$
\end{itemize}
These "abnormal" values might in fact be totally legit in some cases. in that
case the analyzer
-will rank the mutation rather poorly, which will lead to this tree path not
being very likely to be developed further more.
+will rank the mutation rather poorly, which will lead to this tree path not
being likely to be developed further more.
\\*
\paragraph{SQL handling}
-All the SQL statements are generated within the code. This means that the data
concerning the current and future state of the mutations have to be very
precise. Otherwise, the SQL statement is very likely to fail. Sadly, since
SchemaFuzz only supports postgreSQL, the implemented syntax follow the one of
postgres
-DBMS. This is already a very big axis for future improvements and will be
detailed in the dedicated section.
-The statement is built to target the row as precisely as possible, meaning
that it uses all of the non fuzzed values from the row to avoid updating other
row accidentally. Only the types that can possibly be fuzzed will be used in
the building of the SQL statement. Since this part of the code is very delicate
in the sense that it highly depends on an arbitrary large pool of variables
from various types it is a good bug provider.
+All the SQL statements are generated within the code. This means that the data
concerning the current and future state of the mutations have to be precise.
Otherwise, the SQL statement is likely to fail. Sadly, since SchemaFuzz only
supports postgreSQL, the implemented syntax follow the one of postgres
+DBMS. This is already a big axis for future improvements and will be
detailed in the dedicated section.
+The statement is built to target the row as precisely as possible, meaning
that it uses all of the non fuzzed values from the row to avoid updating other
row accidentally. Only the types that can possibly be fuzzed will be used in
the building of the SQL statement. Since this part of the code is delicate in
the sense that it highly depends on an arbitrary large pool of variables from
various types it is a good bug provider.
\paragraph{Injecting} :
The injection process sends the built statement to the DBMS so that the
modification can be operated. After the execution of the query, depending of
the output of the injection (one modification, several modifications, transfer)
informations are updated so that they can match the database state after the
modification. If the modification failed, no trace of this mutation is kept, it
is erased and running goes on like nothing happened.
@@ -209,7 +194,7 @@ The injection process sends the built statement to the DBMS
so that the modifica
The mutation transfer is a special case of a modification being applied to the
database.
It is triggered when the value that was supposed to be fuzzed is under the
influence of a FKC as the child.
In the case a FKC (In CASCADE mode), only the father can be changed, which
also triggers the same modification on all of his children. The algorithm then
"transfers" the modification from the original mutation to its father.
-After injecting the transfered mutation, the children mutation is indeed
modified but the modification "splashed" on some parts of the database that was
not meant to be changed.
+After injecting the transfered mutation, the children mutation is modified but
the modification cascades on some parts of the database that was not meant to
be changed.
Hopefully, this does not impact the life of the algorithm until this mutation
is reverted (see next paragraph).
\paragraph{Do/Undo routine} :
The Do/Undo mechanism is at the center of this software. Its behavior is
crucial for the execution and will have a strong impact on the coherence of the
data nested in the code or inside the target database throughout the runtime.
@@ -219,6 +204,23 @@ Reverting mutations is the key to flawlessly shifting the
current position in th
The case of the transfered mutation is no exception to this. In this case, the
mutation applied changes on an unknown number of fields in the database. But,
the FKC still bounds all the children to their father at this point (this is
always the case unless this software is not used as intended).
Changing the father's field value back to its original state will splash the
original values back on all the children.
This mechanism might trigger failing mutations in some cases (usually
mutations following a transfer). This issue will be addressed in the known
issues section.
+
+ \subsubsection{Choosing pattern}
+For each iteration of the main loop, a modification has to be picked up as the
next step in the fuzzing process. This is done by considering the current state
of the tree.
+Three parallel code paths can be triggered from this point.
+ \begin{itemize}
+ \item{Continue on the current branch of the
tree (triggered if the last mutation scored better than its parent)}
+ \item{Pick an existing branch in the tree and
grow it (triggered if the last mutation scored worse than its parent on a 50/50
chance with the next bullet)}
+ \item{Start a new branch (triggered if the last
mutation scored worse than its parent on a 50/50 chance with the previous
bullet)}
+ \end{itemize}
+
+ \bigskip
+ \begin{figure}[ht!]
+
\includegraphics[width=\textwidth]{pickingPaternDiagram.pdf}
+ \caption{Diagram that shows, for each iteration of the
main loop, the mutation is }
+ \end{figure}
+
+ \bigskip
\subsubsection{Tree Based data structure}
All the mutations that are injected at least once in the course of the
execution of this software will be stored properly in a tree data structure.
Having such a data structure makes parent-children relations between mutations
possible. The tree follows the traditional definition of the a n-ary
algorithmic tree.
It is made of nodes (mutations) including a root (first mutation to be
processed on a field selected randomly in the database)
@@ -226,33 +228,33 @@ Each node has a number of children that depends on the
ranking its mutation and
\paragraph{Weight}
Weighting the nodes is an important part of the runtime. Each mutation has a
weight that is equal to the analyzer's output. This value reflects the
mutation's value. If it had an interesting impact on the target program
behavior (if it triggered new bugs or uncommon code paths) than this value is
high and vice-versa. The weight is then used as a mean of determining the
upcoming modification. The chance that a mutation gets a child is directly
proportional to its weight.
This value currently isn't biased by any other parameter, but this might
change in the future.
- \paragraph{Path} %% changer la frase sur resolve
+ \paragraph{Path}
Since the weighting of the mutation allows to go back to previous more
interesting mutations,
-there is a need for a path finder mechanism. Concretely, this routines
resolves the nodes that separate nodes A and B in the tree. A and B might be
children and parent but can also belong to completely different branches. This
path is then given to the do/undo routine that processes back the modifications
to set the database up in the required state for the upcoming mutation.
+there is a need for a path finder mechanism. In practice, this routine
resolves the chain of node that separates two nodes in the tree. This is done
by, from both nodes, going in the direction of the root until a common ancestor
is found. Fusing the lists of both chains results in creating the full path
between the two nodes. The path is then used when the main loop goes through
the undo mechanism. Undoing from mutation A to mutation B is implemented as
undoing every mutation between A and B
\bigskip
\begin{figure}[h!]
\centering
\includegraphics[width=\textwidth]{CommonAncestorDiagram.pdf}
-\caption{Objects returned by the meta data extraction routine.}
+\caption{Example of path between two nodes in the tree}
\end{figure}
\bigskip
\subsubsection{The analyzer}
-Analyzing the output of the target program is another critical part of
SchemaFuzz. The analyzer parses in the stack trace of the target software's
execution to try measuring its interest. The main criteria that defines a
mutation interest is its proximity to previously parsed stack traces. The more
distance between the new mutation and the old ones, the better the ranking.
%%enlever les mots definis plus bas.
+Analyzing the output of the target program is another critical part of
SchemaFuzz. The analyzer parses in the stack trace of the target software's
execution in order measure how interesting the output of the execution was.
Since crashes and unexpected behavior from the target software is what the tool
is triggering, it is the main criteria of a valuable mutation. A stack trace is
a text block structured to present all the information related to a crash
during a software's execution. The [...]
\paragraph{Stack Trace Parser}
The stack trace parser is a separate Bash script that processes stack traces
generated by the GDB C language debugger and stores all the relevant
informations (function's name, line number, file name) into a Java object. The
parser also generates as a secondary job a human readable file for each
mutation that synthesizes the stack trace values as well as additional
interesting information useful for other mechanisms (that also require
parsing). These additional informations include the p [...]
\paragraph{Hashing}
-In order to be used in the clustering algorithm, the stack trace of a mutation
has to be hashed.
+The clustering algorithm used to compute the crash rank take a triplet of
numerical values as an input.Therefore, the stack trace of a mutation has to be
hashed into a triplet of numerical values. This set of value is used as a
representation of the original stack trace object.
Hashing is usually defined as follows :
\begin{quotation}
"A hash value (or simply hash), also called a message digest, is a number
generated from a string of text. The hash is substantially smaller than the
text itself, and is generated by a formula in such a way that it is extremely
unlikely that some other text will produce the same hash value."
\end{quotation}
-In the present case, we used a different approach. Since proximity between two
stack traces is the key to a relevant ranking, it is mandatory to have a
hashing function that preserves the proximity of the two strings.
+In the present case, we used a different approach. Since proximity between two
stack traces is the key to a relevant ranking, it is mandatory to have a
hashing function that preserves the proximity of two strings.
In that regards, we implemented a version of the Levenshtein Distance
algorithm.
-This algorithm can roughly be explain by the following :
+This algorithm can be explained by the following statement:
\begin{quotation}
"The Levenshtein distance between two words is the minimum number of
single-character edits (insertions, deletions or substitutions) required to
change one word into the other."
\end{quotation}
@@ -271,35 +273,72 @@ This algorithm can roughly be explain by the following :
The distance for this example is $2\div8\times100$
-After hashing the file name and the function name into numerical values trough
Levenshtein distance, we are creating a triplet the fully (but not fully
accurately yet) represents the stack trace that is being parsed. This triplet
will be used in the clustering method.
-
+After hashing the file name and the function name into numerical values trough
Levenshtein distance, the analyzer creates the triplet that numerically
represents the stack trace being parsed. This triplet will be used in the
clustering method detailed in the following paragraph.
+It is interesting to note that this triplet is not the most accurate
representation of a stack trace. The analyzer will be improved in the future is
that regard.
\paragraph{The Scoring mechanism}
-The "score" (or rank) of a mutation is a numerical value that reflects how
interesting the outcome was. Crashes and unexpected behavior to raise this
value whereas no crash tend to lower it. This value is calculated through a
modified version of a clustering method that computes an n-uplet into a integer
depending on the sum of the Euclidean distances from the n-uplet to the
existing centroids (groups of mutation's n-uplets that were already processed).
-This value is then set as the mutation's rank and used as a mean of choosing
the upcoming mutation.
+The "score" (or rank) of a mutation is a numerical value that reflects how
interesting the outcome was. Crashes and unexpected behavior are what makes a
mutation valuable since it indicates a wrongly implemented code piece in the
target source in most cases. This value is calculated through a modified
version of a clustering method.
+This clustering mechanism runs as follows:
+ \begin{itemize}
+ \item{Represent the triplets in a 3 dimensional space}
+ \item{Create clusters that includes most similar triplets}
+ \item{Calculate the centroid of each cluster}
+ \item{Calculate the Euclidean distance between the current mutation's
triplet and all the centroids}
+ \item{Add up all the distances generated by the last step into a single
value}
+ \end{itemize}
+
+The centroid of a cluster is the triplet of values that define the center of a
cluster.
+the Euclidean distance is defined as
+ \begin{quotation}
+ In mathematics, the Euclidean distance or Euclidean metric is the
"ordinary" straight-line distance between two points in Euclidean space
+ \end{quotation}
+
+If a triplet represents a unique crash it will be placed far away in the
Euclidean space. This induces that the sum of the Euclidean distances to the
centroids will be a high value compared to a common crash. This sum is then
used as the "score" of the mutation.
+
+Mutations that do note trigger any crash result in having a null score.
Therefore, the side of the tree they are in has a lower statistical chance of
being chosen for further exploration.
+For a more concrete view of what the analyzer outputs, please refer to the
Result and Example section.
\begin{figure} [h!]
\includegraphics[width=\textwidth]{Scoring.png}
\end{figure}
\subsection{Known issues}
About one mutation out of 15 will fail for invalid reasons.
\subsubsection{Context Coherence}
-A significant amount of the failing mutations do so because of the transfer
mechanism. As said in the dedicated section, this mechanism applies more than
one change to the database (Potentially the whole database). In specific case,
this property can become problematic.
-More specifically, when the main loop identifies a mutation's child as the
upcoming mutation and its parent row has been splashed with the effect of a
transfer. In this case, the data embedded in the schemaFuzz data structure may
not match the data that are actually in the database, this delta will likely
induce a wrongly designed SQL statement that will result in a SQL failure
(meaning that 0 row were updated by the statement).
+A significant amount of the failing mutations do so because of the mutation
transfer mechanism. As said in the dedicated section, this mechanism applies
more than one change to the database (potentially the whole database). In
specific cases, this property can become problematic.
+More specifically, when the main loop chooses the next mutation and its parent
has been the subject of a transfer. In this case, the data embedded in the
schemaFuzz data structure may not match the data that are present in the
database, this delta may induce a wrong SQL statement that will result in a SQL
error (in practice, the DBMS indicates that 0 rows were updated by the
statement).
\subsubsection{Foreign Key constraints}
For a reason that is not yet clear, some of the implied FKC of the target
database can't be properly set to CASCADE mode. This result in a FKC error
(mutation fails but the program can carry on)
\subsubsection{Tests}
-Besides the test suit written by the SchemaSpy team for their own tool (still
implemented in SchemaFuzz for the meta data extraction), the tests for this
project are very poor. Their are only very few of them and their utility is
debatable. This is due to the lack of experience in that regard of the main
developer. Obviously, we are very well aware of this being a really severe flaw
in this project and will be one of the main future improvements.
-This big lack of good maintenance equipment might also explain some of the
silent and non silent bugs that still remain in the code to this day.
+Due to a lack of time and a omission in the project planning, this project's
test suit is not yet complete. In its current state, the test suit includes the
tests written for the meta data extraction routine as well as a bundle of unit
tests that cover the following points :
+ \begin{itemize}
+ \item{instantiation of the Mutation class}
+ \item{Creation of the modification possibilities}
+ \item{the do/Undo routine}
+ \item{Uniformity of the tree weighting}
+ \end{itemize}
+
+The following list details the tests that will be implemented in future
releases by order of importance.
+
+ \begin{itemize}
+ \item{Integration tests}
+ \item{Regression tests}
+ \item{More complete and specific Unit tests.}
+ \item{Performance tests}
+ \end{itemize}
\subsubsection{Code Quality}
-We are well aware that this tool's source code is of debatable quality. This
fact induces the bugs and unexpected behaviors discussed earlier on some
components of this program.
-The following points constitute the main flaw of the source code:
+The code in its current state is still in beta. This means that the code will
be the subject of structural and syntax changes. The following list contains
the major aspects of these changes
+
\begin{itemize}
- \item Hard to maintain. The code is not optimized
either in term of size or efficiency. Bad coding habits
tend to make it rather weak and unstable to context changes.
- \item Structure is not intuitive. The main loop of the
program lacks a good structure.
+ \item{Code structure}
+ \item{Code concision}
+ \item{Code style. More precisely, updating code pieces
that cointain bad coding habits}
\end{itemize}
-
+For example, the following code:
+$if(myVariable == 0)$
+should be changed to:
+$if(0 == myVariable)$
+to avoid unwanted affectation in the case of the omission of an $=$ sign.
\clearpage
\section{Results and examples}
@@ -394,21 +433,21 @@ This section will provide more insights on the future
features that might/may/wi
Any suggestion will be greatly appreciated as long as it is relevant. All the
relevant information regarding the contributions are detailed in the so called
section.
\subsection{General Report}
-In its future state, SchemaFuzz will generate a synthesized report concerning
the overall execution of the tool (which it does not do right now). This
general report will primarily contain the most "interesting" mutations (meaning
the mutations with the highest score mark) for the whole run.
-A more advanced version of this report would also take into account the code
coverage rate for each mutation and execute a last clustering round at the end
of the execution to generate a "global" score that would represent the global
value of each mutations.
+In its future state, SchemaFuzz will generate a synthesized report concerning
the overall execution of the tool. This general report will primarily contain
the most "interesting" mutations (meaning the mutations with the highest score
mark) for the whole run.
+A more advanced version of this report would also take into account the code
coverage ratio for each mutation and execute a last clustering round at the end
of the execution to generate a "global" score that would represent the global
value of each mutations.
\subsection{Code coverage}
-We are considering changing or simply adding code coverage in the clustering
method as a parameters.Not only would this increase the accuracy of the scoring
but also increase the accuracy of the "type" of each mutation. To this day,
this tool does not make a concrete difference in terms of scoring or
information generating (reports) between a mutation with a new stack trace in a
very common code path and a very common stack trace in a very rarely triggered
code path.
+We are considering changing or simply adding code coverage in the clustering
method as a parameters. Not only would this increase the accuracy of the
scoring but also give more detail on what the mutation triggered in the target
software's code therefore helping locate the origin of the crash. By adding
code coverage this tool could make a concrete difference in terms of scoring
and informations being generated in the reports between a mutation with a new
stack trace in a common code pat [...]
\subsection{Data type Pre-analyzing}
This idea for this feature to be is to implement some kind of "auto learning"
mechanism.
To be more precise, this routine is meant to performed a statistical analysis
on a representative portion database's content. This analysis would provide the
rest of the program the most common values encountered for each field. More
generically, this would allow the software to have a global view over the
format of the data that the database holds.
-Such global understanding of the content format is very interesting to make
the modifications possibilities more relevant. Indeed, one of the major
limitation of SchemaFuzz is its "blindness".
+Such global understanding of the content format is interesting to make the
modifications possibilities more relevant. Indeed, one of the major limitation
of SchemaFuzz is its "blindness".
That is to say that some of the modifications performed in the course the
execution of the program are irrelevant due to the lack of information on what
is supposed to be stored in this precise field.
-For instance, a field that only holds numerical values that go from 1 to 1000
even if it has enough bits to encode from -32767 to 32767 would have a very low
chance of triggering a crash if this software modifies its value from 10 to 55.
-on the other end, if the software modifies this very same field from 10 to
-12000, then a crash is much more likely to pop up.
+For instance, a field that only holds numerical values that go from 1 to 1000
even if it has enough bits to encode from -32767 to 32767 would have a low
chance of triggering a crash if this software modifies its value from 10 to 55.
+on the other end, if the software modifies this same field from 10 to
-12000, then a crash is much more likely to pop up.
Same principle applies to strings. Suppose a field can encode 10 characters.
-the pre-analysis, detected that, for this field, most of the value were
surnames beginning with the letter "a". Changing this field from "Sylvain" to
"Sylvaim" will probably not be very effective. However, changing this same
field from "Sylvain" to "NULL" might indeed triggered an unexpected behavior.
+the pre-analysis, detected that, for this field, most of the value were
surnames beginning with the letter "a". Changing this field from "Sylvain" to
"Sylvaim" will probably not be effective. However, changing this same field
from "Sylvain" to "NULL" might indeed triggered an unexpected behavior.
This pre-analysis routine would only be executed once at the start of the
execution, right after the meta data extraction. The result of this analysis
will be held by a specific object.
this object's lifespan is equal to the duration of the main loop's execution
(so that every mutation can benefits from the analysis data.)
diff --git a/docs/PersonnalExperience.tex b/docs/PersonnalExperience.tex
index b07036f..add1e07 100644
--- a/docs/PersonnalExperience.tex
+++ b/docs/PersonnalExperience.tex
@@ -1,5 +1,5 @@
-\section{Internship organisation}
+\section{Internship organization}
\subsection{Introduction}
This section is meant to be added to the University version of this
documentation. It will be written as Erwan Ulrich and will focus on the
different aspects of the organization of the project. The following text will
also be written with a more personal and more critical point of view as a mean
of self analyze.
@@ -20,7 +20,13 @@ It is also a personal reminder of what should be improved in
my work habits and
\begin{itemize}
\item{Defining tasks/features as daily/weekly sub goals}
- \item{Improving general project planning} %% bad title.
+ \item{Improving general project planning}
+ \begin{itemize}
+ \item{Include the test writing in the planning as a "real" task}
+ \item{Build the project's code structure beforehand}
+ \item{Decide what approach to use for each component beforehand}
+ \item{Decide for each component what technologies should be
used beforehand}
+ \end{itemize}
\item{Setting up more fluid communication}
\end{itemize}
--
To stop receiving notification emails like this one, please contact
address@hidden
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [GNUnet-SVN] [taler-schemafuzz] branch master updated: rewrote so paragrphs that were unclear,
gnunet <=